Total Invisibles: don't fail assert()s that we tried to add in #433

[thomas-maeder]

See #433

[JoshuaGreen]

Before we go too crazy here, I'd like to emphasize some points.  My

assert(en_passant_was_multistep_played(entry->u.ep_capture_potential.ply));

assertion (which when paired with a correction to en_passant_was_multistep_played strengthened @thomas-maeder's

assert(en_passant_top[entry->u.ep_capture_potential.ply]>0);

assertion) captured what seems like a logical assumption, namely that we should never try to remove an en passant square that we didn't actually add.  On the other hand, my

assert(entry->u.ep_capture_potential.ply>=nbply);

assertion primarily reflects a limitation of our en passant infrastructure.  It's conceivable that some solving requires us to manipulate rangers earlier than nbply, and in that case it would be the en passant infrastructure that would have to be appropriately updated to accommodate that.

[thomas-maeder]

@JoshuaGreen I think this issue is obsolete. Do you agree?