set_hh_evalue() length bins cause DISOPRED profiles to diverge too far from standalone

[DanBuchan]

run_hhblits_legacy_blast.py sets the hhblits evalue dependent on seq length size. But this markedly changes the profiles for very long sequences and some methods can have noticeably differing results.

Best fix would be to lose this function. Instead kill HHblits after it exceeds some acceptable runtime and then fail over to a psiblast run as per usual.