chore(nnue): promote inc2 net (gen3) — depth-8 self-play warm-start

10 epochs warm-started from gen2v2/current with 6000 depth-8 self-play
games (697k positions) mixed 70/30 with preserved gen2v2 data. Loss
0.0174 → 0.0143 over 10 epochs, smooth monotonic decrease. A/B eval
spot-check vs predecessor shows agreement on tactical and middlegame
positions, opening-preference shift in startpos (b1c3 → e2e4), no
broken behavior.

Author: Inuway

.gitignore

@@ -4,4 +4,5 @@
 # NNUE training artifacts
 nets/*-data.bin
 nets/*-epoch*.nnue
+training*.log