Implement and tune Lazy SMP search

[nook0110]

Stacked on #44.

Summary:

• add UCI Threads support with Lazy SMP workers sharing a clustered lock-free transposition table
• make TT capacity configurable and back large tables with mmap plus transparent huge-page advice
• expose RFP, NMP, IIR, and LMR tuning options and bake the validated search defaults
• add automated coordinate tuning and reproducible strength-evidence checks

Validation:

• 130/130 enabled regression tests passed
• ThreadSanitizer 100-ply Threads=8 bestmove/ponder stress passed
• tuned vs default: 275-190-303 over 768 clean paired games, +38.6 Elo, LLR 3.549 for H0=0 / H1=+10
• 256 GiB TT smoke observed 257 process threads and a legal best move

The long WinAtChess test and one pre-existing disabled NMP fixture were not run.