A pretty much standard setup for adding crowd sound is two cardioids at the outside edges of the stage, facing away from the stage and towards the audience. Substituting hypers could be a good option if the PA is closer to the reduced sensitivity regions of the hyper pattern and you aren't riding the fader.. or if that's just what you happen to have on-hand. But your problem is the quality of reverberant sound of the PA when the aud mics are up in level, so in combination with auto-ducking or your manual fader riding which lowers the level of the aud mics during the music, you might instead try intentionally getting more direct PA sound into your audience mic(s) if doing so makes the PA less 'overly reverberant sounding'. So the omni suggestion or repositioning your directional mic to get more PA may be good advice. In that case the automatic ducking or manual fader riding adjusts both the level of crowd sound and the clear & direct sound picked up from the PA as well, instead of turning up the room sound with too much PA reverberance. But that only works if you can get a clean PA vocal sound with the aud mic level up as much as you need it between songs.
Really this is two seperate issues. One is the over reverberance of the PA when the aud feed is up in level, the other is semi-automating the fader riding. You'll probably get the best results from optimizing both seperately. You might be able to reduce the reverberant PA effect with careful ducking settings, but that will be trickier to avoid 'pumping'. I realize I'm contradicting myself somewhat by initally suggesting ducking as the solution to your reverberance problem, and that might work, but on thinking it through, I think this likely is a better approach.
If you can get the reverberance taken care of with mic pattern and position, audible pumping of the applause in the ducking setup should be pretty easy to eliminate by using appropriate attack and release times. Set your compression ratio to achieve that -10dB or so of gain reduction above the side-chain threshold or whatever amount works when manually riding the fader and tweak the treshold point and attack/release times to achieve a smooth fade-in and out of the applause. Using it to cut the excess vocal reverbrance is tricker, because you'll need careful settings that act fast enough to lower the level as soon as the vocal cuts in, but doesn't 'pump' the audience sound.