As mentioned by the others, "safety pair" means just one microphone pair used in isolation without the others will produce an acceptable recording on its own. There are two motivations behind that as I see it: 1) A way to salvage the recording if/when something goes wrong with the other channels. 2) Retaining a "known good" 2-channel microphone configuration reference setup for comparison with the "whole enchilada OMT arrangement" in order to make sure the OMT complications are worthwhile. This second aspect can obviously be achieved with a second recording setup (two extra microphones and recording channels), but might also be a 2-channel subset of the OMT setup.
Usually "safety pair" in the first sense implies that pair is setup in a typical stereo-pair configuration. But that needn't necessarily be the case. How much insurance do you need to feel comfortable? Quite often I find that all pairs making up the OMT can be considered "safety pairs" in this way to different degrees, differing in how much work is needed afterwards to mold it into something listenable on its own, and what level of quality is acceptable for such a situation. It's dependent on your tolerance for risk, your familiarity with the setup, how oddball you want to push things, the number of channels you want to run, and how comfortable you feel with the mixing side of things. As the channel count increases, redundancy and post-manipulation possibilities also increase. A 3' spaced omni pair can work fine on its own and of course an X/Y or M/S pair can too. As for myself, I'd not hesitate to make a Mid/Side re-adjustment to "widen-up" a narrow-angled X/Y "safety-pair" in order to get it sounding as good as possible. Yet as long as it was setup correctly with close coincidence, I have little concern over dialing in enough additional Side for a 50-degree PAS X/Y pair to sound as wide as a 90-degree X/Y pair. The fact that the 50-degree pair was presumably Pointed At Stacks is an advantage which I suspect might result in better sound after such an adjustment (this would make for a very interesting comp!). To my way of thinking, when recording from a distance directness is more primary and cannot be recovered, whereas width is easier to manipulate, less ephemeral, and carries somewhat less perceptual importance. This is a set of compromise tradeoffs of a similar nature to the ones I'll mention below with regards to near-spaced configurations.
Actually I also consider center front-facing / rear-facing microphones as a "safety pair" in this sense, and have used them as such when I've had problems which rendered the other channels unusable. That pushes this first aspect of "safety pair", but in this way even just a singular forward-facing center microphone serves as a "mono safety". A mono recording may be preferable for some folks and some recordings, and the all important direct/reverberant balance can still be adjusted somewhat via the balance between the front and rear-facing microphones. It's sort of like varying the polar pattern for a mono pickup from that of the forward-facing microphone through wider patterns to omni (and onward to increasingly rear-facing directionality until reaching the pattern of the rear-facing microphone if the level of the rear channel were to be brought up higher than the front - unlikely in practice, but entirely possible). This variation of the center forward facing microphone pickup pattern is one way of thinking about what mixing in various amounts of a single rear-facing microphone does, and ignoring the spacing between front and back-facing microphones, is similar to how a dual diaphragm microphone with electrically switched patterns works. But it doesn't have to be mono. We can do the Mid/Side stereoization trick of using the rear-facing channel as Side in a Mid/Side converter to derive a stereophonic quality, even if that doesn't incorporate left/right directional imaging but rather stereo width, spaciousness and depth. That avoids the odd Left/Right lopsided stereo output which would result from the front microphone channel routed left and the rear microphone routed right, which is typically how monitoring that pair directly sounds - although believe it or not I have a few recordings which actually work well that way.
The second, known-good setup "safety pair" for comparison thing is super valuable for making truthful assessments of what is working and what isn't. And it's similarly valuable for confirming forward progress in determining which OMT variation is satisfactory, and whether that remains sufficient for you or if you want to push it further. I've written a lot about this comparison aspect in these threads and don't want to repeat myself too much, but I will say that listening closely, hearing the differences for yourself, identifying what is working better and what isn't, and figuring out what to correctly attribute that to is key. It's great if the "known good" 2-channel arrangement dovetails nicely with OMT so that you don't need to run redundant channels, but I caution folks from getting overly saddled with trying to "build up OMT" around a favorite near-spaced configuration. One way of thinking about near-spaced 2-channel stereo configurations is that they work well partly because they represent various "best-compromises" between 2-channel coincident and 2-channel spaced configurations. I hear those compromises inherent in near-spaced configurations as very appropriate, good and necessary, yet neither subtle nor minor once exposed by their absence. OMT is in effect a way of working around those compromises, and us most effective when it breaks free of them.
Yes OMT introduces alternate compromises in increased channel interactions and phase relationships (and appropriate mixing requirements) and these can be as egregious as those of coincident, near-spaced or widely-spaced 2-channel configurations or even worse. Yet when dealt with appropriately I find the compromises can be made to be considerably less significant, such that to my ear at least, a good OMT arrangement is often better suited to the recording and reproduction of live-performance music, which is in itself an oddball recording endeavour!
It is difficult to determine what is really most appropriate and what might be better than our current conception of good and best. A lot of that comes from training one's ear as well as re-calibrating one's expectations over time. As tapers, we've made and listened to years, decades-worth of live recordings. We know what works well in a 2-microphone/2-channel paradigm, what sounds good to us, and our brains are trained to hear recordings which convey appropriate traits as good. In this context we even hear necessary compromises as being good rather than inherent to 2-channel recording. We can add a second pair of mics and it may sound better, yet we are judging that based on all our ingrained expectations. It's hard to know what could be even better, and at the same time we are not well attuned to the new compromises our brains are less familiar with, which escape notice until we become familiar enough with the new paradigm to begin to recognize them.
I can now listen to a recording and be completely satisfied with it in every way, a great recording, yet know that it could be still better (not necessarily made by me). Center solidity and ambient envelopment are in this category for me. We are so used to these things being a trade off against each other, sort of a balancing act, that we accept compromises in these things without even being aware of it. Once the compromising constraints tying them together are eliminated and exploited we can recognize that it sounds sounds better, yet that still doesn't re-calibrate our brain to make that our "new normal" for a long time. It's almost like we get surprised by how good it sounds when we do come across it again. I aim to re-calibrate my unconscious expectations toward more "and's" between qualities and away from unstated "or's" - center solidity that surprises me in its solidity in combination with holographic directional stereo imaging, in combination with deep and wide envelopment along with the most appropriate direct/reverberant balance, timbre, and presence. I want all that, even though I know I don't fully recognize what it can be if fully realized.
All this OMT stuff is just a method to try and get closer to that.