Release CE — stage-400 — v0.51.107 — 8-PR low-risk batch (pin-cap getter rename + upload-turn dedup + active-run repair + KaTeX incremental + profile default model + French locale + update-check error surfacing + update apply path)

[nesquena-hermes]

Release CE — stage-400 — v0.51.107 — 8-PR low-risk batch

Pre-merge gate: ✅ JS+Python syntax, ✅ no merge markers, ✅ no high-risk paths touched, ✅ no Docker / no new concurrency / no CJK escapes, ✅ full pytest 6254 passed (0 failures, after fixing 2 callsite tests that asserted the old function name #2718 renamed), ✅ Opus pre-merge composition review = SHIP with all 4 cross-PR concerns verified clean against master.

PRs in this stage (smallest → largest)

• fix: rename _pinnedSessionsLimit() to _getPinnedSessionsLimit() to avoid naming collision with config var #2718 by @eslicarrillo — Rename _pinnedSessionsLimit() → _getPinnedSessionsLimit() (follow-up to v0.51.105 feat: make pinned session limit configurable #2700, matches _get*() convention). All 3 callsites updated; window._pinnedSessionsLimit config var deliberately preserved
• fix(chat): dedupe uploaded-file optimistic user turns #2723 by @ai-ag2026 — Dedupe uploaded-file user turns when optimistic bubble = plain text but persisted pending turn has [Attached files: ...] suffix
• fix(session): treat active runs as live during repair #2721 by @ai-ag2026 — During session repair on restart, guard the restart-stale prune with active-stream check so live in-flight runs survive
• fix: render streamed math incrementally #2710 by @Michaelyklam — Render streamed KaTeX math incrementally (no prose-with-math flash before final render)
• fix(model): prefer profile default on fresh boot #2709 by @starship-s — Fresh-boot: prefer active profile's configured default model over static HTML placeholder. Behavior note: now wipes persisted user pick on refresh (intentional, ratified by tests/test_model_default_boot_precedence.py)
• Complete French (fr) locale: +93 missing keys (88.8% → 96.7%) #2722 by @victorwhale — Complete French (fr) locale: +93 missing keys, 88.8% → 96.7% coverage
• fix: surface update check fetch errors #2717 by @ai-ag2026 — Surface update-check fetch errors instead of failing silently
• fix: apply release update target #2719 by @ai-ag2026 — Apply release-update target on explicit "Check for updates" click (clears sessionStorage check-once stamp + forces banner re-evaluation)

Cross-PR composition review (Opus verified, all clean against master)

Concern	Evidence
(A) #2718 rename × v0.51.105 #2700 callsites	All 3 function callsites in static/sessions.js updated; window._pinnedSessionsLimit config var deliberately preserved (different identifier)
(B) #2710 KaTeX × #2723 user-turn dedup	Role-gated (dedup is user-only via _sameTranscriptMessage user branch); KaTeX path is assistant-stream-only. No overlap
(C) #2709 boot-default × v0.51.104 #2674 new-chat resync	Boot IIFE is one-shot; newSession runs later on user action with server-resolved model. Sequence: boot → profile default fills dropdown → user click → server response → #2674 resync runs uncontested
(D) #2717 error surfacing × #2719 apply path	Explicit "Check for updates" sends ?force=1; check_for_updates(force=True) bypasses the CACHE_TTL = 1800 short-circuit, so cached error result can't block re-evaluation. Apply path independent of _update_cache

Mid-batch test correction

Two existing tests asserted the OLD function name _pinnedSessionsLimit(). The #2718 PR didn't update them. I patched both as part of the stamp commit:

• tests/test_configurable_pinned_sessions_limit.py:54-55 → _getPinnedSessionsLimit()
• tests/test_issue2508_session_pin_cap.py:78-79 → _getPinnedSessionsLimit()

All 6 affected tests pass after the patch.

Invariants preserved

• v0.51.96 fix(session): use second-level timestamp granularity in legacy dedup key #2620 second-level timestamp dedup ✓
• v0.51.99 fix: cap CLI session sidebar state scans #2647 capped CLI sidebar candidate window ✓
• v0.51.99 fix: dedupe replayed context tail after compression #2651 replay-tail dedup ✓
• v0.51.100 fix(session): lazily retry run-journal recovery so the interrupted-turn marker self-heals #2615 lazy journal recovery ✓
• v0.51.100 Sync session list from server events #2637 SSE session events ✓
• v0.51.103 fix(plugins): distinguish exclusive/provider activation in Settings panel #2663 plugins exclusive activation ✓
• v0.51.104 Improve session-event reconnects and cron run expansion #2671 SSE reconnect jitter ✓
• v0.51.104 Fix new chat default model provider sync #2674 new-chat model picker provider resync ✓ (verified non-conflicting with fix(model): prefer profile default on fresh boot #2709)
• v0.51.104 fix(ui): invalidate transcript cache on same-count content changes #2692 transcript cache content-hash signature ✓
• v0.51.105 feat: make pinned session limit configurable #2700 configurable pin cap (range 1–99, default 3) ✓ (fix: rename _pinnedSessionsLimit() to _getPinnedSessionsLimit() to avoid naming collision with config var #2718 renames accessor only)
• v0.51.106 fix(session): dedupe restamped state rows in display merge #2686 restamped state.db replay dedup ✓
• v0.51.106 fix(api): deduplicate context messages to prevent agent from seeing duplicates #2705 _deduplicate_context_messages() ✓
• v0.51.106 fix(streaming): prevent exponential _partial message accumulation #2704 _message_identity tuple for empty _partial ✓