fix(stats): correct harness pass count 445->446 (flaked local summary)

The v0.6.0 publish shipped STATS.json with harness.pass:445 - a transient perf-gate flake on a slow local FS. CI ran a clean 446/446, so build_stats --check and check_stats_drift saw the committed STATS as stale and went red. Regenerated STATS.json/STATS.md from a clean 446/446 run.

Author: elad-reznikov

validation/STATS.json

@@ -171,7 +171,7 @@
       "recall_pct": 0.6364,
       "wilson95_lb": 0.4295
     },
-    "pass": 445,
+    "pass": 446,
     "sections": 99,
     "total": 446
   },

validation/STATS.md

@@ -3,7 +3,7 @@
 *Generated by `tools/build_stats.py` from the committed validation records + the harness summary.*
 *Do not hand-edit — every figure self-derives. Re-run `python tools/build_stats.py`.*
 
-**Harness:** 445/446 checks PASS across 99 oracle sections.
+**Harness:** 446/446 checks PASS across 99 oracle sections.
 **External metamorphic oracle (§26):** recall 14/22 = 64% (Wilson95 LB 43%); metamorphic-FP **0** (the hard gate). Recall is INDICATIVE — §18 stays the headline.
 
 ## Mirror Test (core-output coverage vs the project's own docs)