Puzzling Ending

HERE is the ending of a game that has not made its way into ChessBase's Mega database.

White to make his 38th move in Joseph Blackburne - Louis van Vliet & James Manlove, London 1893

*****
*****
*****
*****
*****
*****
*****
*****
Black is a pawn up and, in my opinion, has the better minor piece (the pawns have little flexibility of movement, and two white pawns are fixed on dark squares - the same colour as the bishop). But the presence of rooks favours White, as he has the bishop, and it is far from obvious, at least to me, how Black is to make progress. For example, an exchange of minor pieces to remove the blockader of Black's passed pawn would result in a rook-and-pawn ending, with definite drawing chances for White.
Stockfish10 reckons Black's advantage is worth just under half a pawn. Komodo9 rates Black's chances as higher than this, but only by about a fifth of a pawn.
Notes in italics (algebraicised where appropriate) are by Tartakower & du Mont from 500 Master Games Of Chess.
They introduced the game thusly: The salient feature of this fine game is the clever manner in which the consultants gradually succeed in wresting more and more territory from their great opponent, the endgame in particular being of classic beauty.
38.Rc2?
This move passes unremarked, but allowing the rooks to come off virtually guarantees White's defeat. Perhaps there is some feature of the position I am missing, but I find it almost impossible to understand why Blackburne would play in this way, and why Tartakower would not call him on it.
The engines suggest 38.Re1, which stops Black's rook from invading.
38...Kd7
Accepting the challenge of a duel between knight and bishop.
Without actually saying so, Tartakower implies that allowing rooks to come off is a brave decision by Black (presumably on the ground that a bishop is often better than a knight in an ending with pawns on both sides of the board). However, Black did not have much choice, eg 38...Ra6? 39.Rc7 allows White right back into the game.
39.Rxc6
If White had changed his mind about exchanging with, say, 39.Re2, White's rook would have invaded with 39...Rc1.
39...Kxc6 40.Bg7 Nc5+ 41.Ke3
White's king would be nearer his vulnerable queenside pawns with 41.Kd2, but then Black's king would invade with 41...Ne6 42.Be5 d4 43.Kd3 Kd5 with ...Nc5+ to come.
41...Ne6
Good enough, but even stronger is 41...a4, when White will eventually lose both his queenside pawns for just one of Black's.
42.Be5 Kc5 43.Bf6
Or 43.Kd3 d4 44.Bb8 Kd5 etc.
43...d4+ 44.Kf3
The king prefers to remain the guardian of the f4 pawn.
44...Kb5 45.Be5 Kc5 46.Bb8 Kd5 47.Ba7
The bishop has to leave the useful diagonal. If 47.Be5, 47...Nxg5+ etc.
47...d3 48.Ke3 Nxf4!
The beginning of the end.
49.Bb6 a4 50.Bd8
Not 50.bxa4 Kc4 etc, nor 50.Kxf4 d2.
50...axb3 51.axb3 Ne6 52.Bb6 Nc5
Well thought out, for in the king's endgame Black's passed pawn ensures victory.
53.Bxc5 Kxc5 54.Kxd3 Kd5 55.Ke3 Ke5 56.h3 f4+
And Black wins.