The possible 23.Nxe6? looks like a gimme, but the attacker would be pinned by the reply 23...Qd7. "Can this move be pinned/skewered/forked on it's new square?" is a mighty good question to ask yourself before/after any move is made.

28.Qxc5+ was certainly playable instead of exchanging queens and pawns. A much different game would result w/queens remaining on board. White would have to defend his backward d-pawn instead of the actual 29.dxc5 to open the file. The queenside exchanges that do occur allow White's pieces to breath, but he's down a piece and the ability to deliver a mating net is greatly reduced.

After further piece exchanges, Black's two minor pieces (an advantage initially resulting from the early sacrificial exchanges on g5) prove better than the lone White rook. It is unusual to see Black's dark-squared bishop play guard duty of his pawn after 43...e5 for 17 moves (like a bad bishop), yet turn up on the winning side. Black was already better at that point, having the more advanced king and pawns. Maintaining the e5-pawn is important, as a sacrificial R x c-pawn is a pointless option w/another pawn available.

I probably would have played 40.f4 but I don't see a way for the White pawns to advance safely on the kingside. 46...Nxb3 is a game winner, creating a decisive passed pawn. The Black knight is just right in this ending.