A bad move, I think.

The Knight is imprisoned there.

He had to defend the knight.

(!) Complicating matters: a very good idea! Black has to be careful: 22...fxe5?! 23.Nf6ch gxf6?? 24.Bh7ch! Kxh7 25.Rxc3 and White wins. Or (worse still) 22...Qxe5? 23.Re3 Qh5ch 24.f3 d5 25.g4! and if 25...Qf7, 26.Kf1 prepares to double rooks on the e-file to which Black has to find an answer. If capturing the e-pawn is insufficient either way, how, then, should Black play here?

But much better was 22...Qxe5+. And I don't see how this helps in getting the kngiht?

Yes, I do believe that 22...Qxe5 was better than this King move, but it would still have been insufficient. The right move is in fact fairly hard to find - certainly it isn't a move that leaps to the eye: 22...Ba6!! The point of this is the half-pin on the White Bishop. If now 23.Bxa6 Qc2ch 24.Ke3 Qxb1. See how the White bishop has been decoyed away from protecting the c2-square - let alone the h7-Knight! 25.Bd3 Qe1ch 26.Kf3 Qd1ch and if 26.Be2 Qd5ch followed by 27....Kxh7. White's best response is probably 23.Rd1 though after 23...Bxd3ch 24.Rhd3 Qxe5ch Black is just about certain to pick up one of the Knights, after which the Queen and 2 Pawns will have the edge over the pair of Rooks.

I'll take this opportunity here to remark what a fantastic game this is: things happening all over the place. There has been no let up since the first moves. While it is true both sides have missed several opportunities, you have to admit they are creating chances for themselves as much as for each other. It is this kind of game that developing players should be playing as much as possible; you would see a vast improvement in their play in no very great interval of time. This is what makes Gambits worthwhile.

23...Ba6 was still good.

This was a simple, straightforward oversight. Black might have been hoping White would take, but he could scarcely have been counting on it. In fact, White could safely have extricated his Knight now (another reason why ...Kh8 was not a good move: it uncovered the f8-square).

Now White is quite busted.

25...Bb7 wins the knight on a8 because it has no safe square to run to.

Why not go back to f3? Black might be interested in a draw by threefold repetition...

28.Ke3 makes sense, here, as if 28...Qe5ch White could put in a draw claim and then play 29.Kd2.

You could have taken a pawn with check (28...Qxa2+).

28. ... Ba6 is interesting too. If 29. BxB Qxa2+ wins the rook. The threat is 29. ... BxB+ 30. KxB Qf5+, picking up either the rook or the knight (whichever one doesn't move on black's 29th move).

I would prefer 30.Kf1, freeing up the e-file for the rook.