Phase 13.49.DF FIX1: HTML rendering fixes (H-1/H-2/H-3) + M.3 lock.

H-2: scope JS filter selectors to '.filter-group [data-status]' /
'.filter-group [data-visual]' — broad selector also matched the 121 feature
<tr> rows (which carry data-status for applyFilters), so clicking a feature
row hijacked filter state.

H-3: sort features by category (stable) before emit in both MD and HTML
paths. FEATURES is in chronological commit order; without the sort the
per-transition header was emitted multiple times per category (FACET 5x,
PROFILE 4x, HIST 4x).

H-1: cannot reproduce in container — same test_results / same
compute_feature_stats produces Verified in both MD and HTML. Committed
HTML was a stale artifact (generated pre-D-L fix when M.1 was failing).
Resolved by regeneration; locked against recurrence by the M.3 extension.

M.3 extension: gate-locks two invariants on the HTML output:
  (a) each category appears as a category-row header exactly once.
  (b) HTML data-status counts agree with MD-computed status counts on the
      same test_results.
Both fail on the original buggy emitter, pass on this one. Same test;
no gate change (1038).

Process note: Sonet51 [X] verdict 2026-05-30 was correct — the 8-reviewer
v1.1 panel (including me) approved without rendering the HTML. Same
failure class as audit PR-3 at the meta level. QRC entry owed.

Predecessor: PHASE_13_49_DF_END.
Gate: 1038 -> 1038 (no test count change; M.3 assertions strictly tighter).

Author: miranov25