Improve FRC correction.

[vondele]

As FRC can not be tested on fishtest, locally tuned and tested:

100k games at STC:
Elo: 2.36 +- 1.07
LOS: 0.999992

Bench: 5714575

[snicolet]

It may also be possible to do the same for lines 482-483 of evaluation.cpp

if (   pos.is_chess960()
   && (s == relative_square(Us, SQ_A1) || s == relative_square(Us, SQ_H1)))
    {
        Direction d = pawn_push(Us) + (file_of(s) == FILE_A ? EAST : WEST);
        if (pos.piece_on(s + d) == make_piece(Us, PAWN))
           score -= !pos.empty(s + d + pawn_push(Us)) ? 4 * make_score(CorneredBishop, CorneredBishop)
                                                      : 3 * make_score(CorneredBishop, CorneredBishop);
   }

Since this code is only used in Chess960 and when the classical eval is called, this is really a corner case :-) and we could even commit such a change directly, as the result in your test is enough to build confidence, IMHO.

[NightlyKing]

Wow, a FRC patch out of the blue. A surprise to be sure - but a welcome one.
Congratulations!

[vondele]

@snicolet I prefer not to change classical eval at this point. It definitely would need testing since the correction should depend on the 'base evaluation', and I think it is not worth resources to do that. I think we should essentially consider the classical eval frozen for now.

[snicolet]

Merged via 329bdbd, congrats :-)