You can use any model from the 30E on upwards for a-a HA, theoretically. In practice, you will see this in midrange models, that is, 100E upwards.
Again, why would I use a-a over a-p? Both modes provide for redundancy, but a-a features load balancing. For instance, with a high number of SSLVPN clients and some other CPU intensive tasks at the same time. I wouldn't expect that when deploying low range models. Given that midrange models of the E or F series are really powerful, I don't see why I should use a-a mode here either.
YMMV, as always.
May I ask you for one favor: please do not rate posts with 1 or 2 stars; this actually reduces reputation. OK if you mean it, but I reckon the "-2...+2" rating scheme is not self-explanatory and your intention was the opposite.
So 1 star="b-s", 2 stars="plain wrong", 3 stars="why rate at all", 4 stars="helpful", 5 stars="cool!"