fix(parallel): drop MinGW emutls thread_local on the worker NA path

[ms609]

Problem

Parallel NA search (nThreads >= 2) intermittently aborts with STATUS_HEAP_CORRUPTION on Windows/MinGW. Serial is unaffected; EW is unaffected.

Root cause

The per-thread scratch in the TBR kernel (ts_tbr.cpp, ts_fitch.cpp) and exact_verify_sweep's optimum cache were function-local static thread_local. On MinGW these resolve via emutls, whose thread_local teardown across std::thread spawn/exit corrupts the heap. EW never trips it (light TLS); the NA path does, because exact_verify_sweep adds a thread_local std::unordered_set plus more scratch.

Fix

• Convert all worker-reachable scratch from static thread_local to plain function-locals — each worker owns its call frame, so they're per-thread-safe with no emutls. Per-clip reallocation measured ≤1.6% on 88-tip data, ~0% typical.
• Move exact_verify_sweep's optimum memoization (evs_false_cache / evs_last_fp) to mutable members on DataSet. Each worker's ds_local has the same per-worker, cross-replicate lifetime the thread_local had, so persistence is unchanged — without emutls.

Verification (clean builds: rm src/*.o; CCACHE_DISABLE=1; --preclean, on this base)

	parallel NA	parallel EW	serial scores	NA perf
base (da0f203)	crashes (iter ~4–8)	200/200	NA=79, EW=78	—
this PR	120/120	200/200	NA=79, EW=78 (bit-identical)	4.15s (cache intact, vs 5.81s cache-disabled)

🤖 Generated with Claude Code