Fix FH_TABLE_COLLECTION makefile wiring; disable LMP censoring during collection builds

FH_TABLE_COLLECTION was referenced in search.c/searchsup.c/chess.h but
had no GNUmakefile wiring to a compiler define -- collateral damage
from the 2026-10-14 PRUNE_VERIFY/PRUNE_LIVE dead-flag cleanup, going by
GNUmakefile.orig's surviving pre-cleanup copy, which still had it.
Every "collection" build in between compiled clean but silently never
included the instrumentation. Restored.

Also: under FH_TABLE_COLLECTION, force fThisMoveLMPPruned = FALSE
instead of calling ShouldLatePrune. ShouldLatePrune is itself partly
driven by uPredictedFH -- a lookup into the CURRENTLY LOADED data.c
table -- so leaving LMP live during a collection run meant the old
(soon-to-be-replaced) table decided which moves were "safe enough" to
go unobserved by the new collection, silently censoring the very
population the new table is supposed to measure. Collection builds now
always try every eligible leftover move.

Found while preparing the NNUE-eval dual-table (alpha+beta FH%)
re-collection run (docs/NNUE.md's eval-swap calibration checklist,
section 3) -- the current tree's collection instrumentation also turns
out to be missing the 5D alpha table's `improve` axis and the entire
beta-side table relative to the original dual_table_run collection;
that restoration is tracked as separate, larger follow-up work.

Co-Authored-By: Claude Sonnet 5 <[email protected]>

0 files changed, 0 insertions, 0 deletions