emrg: result-panel per-session artifact isolation (render gate + switch restore)

[pm25coder]

Summary

Fixes a gap found while testing PR #664 (workspace panel P2 framework, merged as 4c71b25): the result-panel artifact cards were not actually per-session isolated, despite the per-session state table.

The bug

addToolResult(data, sid) correctly stored the record into the per-sid artifactsBySid bucket, but then called renderCard(data) unconditionally — rendering into the shared #result-list DOM. Consequences with multi-session + workspace panel both live on master:

1. A tool_finished event from a background session painted its artifact card into the currently-active session's artifacts pane (cross-session DOM leak).
2. switchSession() never re-rendered cards from the per-sid bucket — switching to a session whose bucket had records showed nothing from that bucket (and stale cards from the previous session lingered).

The fix (result-panel.js)

• addToolResult: store the normalized record to the bucket, render the card only when the event sid is the current session — the same (sid || null) === currentSid gate already used by openFileTab/closeFileTab for the tab strip.
• New renderArtifacts(): rebuilds the artifacts pane from the current session's bucket (mirrors renderTabbar); called from switchSession (so switching restores each session's own cards) and init.
• Card cap stays MAX_ITEMS (50) for the DOM render; the 100-record bucket remains for P3.2 consumption (unchanged from emrg: GUI workspace panel P2 framework — resizable panel + tab bar + per-session tab state (rant 2026-08-11T12:20:35) #664).

Tests

+2 renderer.smoke tests (72 → 74; GUI 193 → 195):

• Background-session tool_finished → 0 cards in the active pane; current-session event → 1 card.
• Register records for two sessions (non-active → bucket only), switch to each → pane shows exactly that session's record, no leakage on switch back.

Verification

• cd emrg/gui && npm test → 195 tests, 194 pass, 0 fail, 1 cancelled (the 1 cancelled is the pre-existing P2 onRecovered timing flake in conn-manager.test.js:488, also cancels on master, passes in isolation — unrelated).
• uv run pytest tests/ → 639 passed, 55 skipped (694; unchanged — no Python changes).
• node --check on touched JS + import + CLI checks: clean.
• Doc-count guard green (Agent.md / README.md / README.cn.md counts 193 → 195).

[argszero]

✅ LGTM — cycle

Reviewed the resolved branch head deb4cc9 (conflict from #665's viewer merge into result-panel.js was already handled on the branch — doc counts synced to 200, both P3 and artifact-isolation test sets kept).

The fix itself is exactly what the P3.2 note called for:

• addToolResult: normalized record stored to the per-sid artifactsBySid bucket; DOM card render gated on (sid || null) === currentSid — same gate openFileTab/closeFileTab already use for the tab strip. Background-session tool_finished no longer paints into the active session's pane.
• renderArtifacts(): rebuilds the artifacts pane from the current session's bucket (mirrors renderTabbar); called from switchSession (each session's own cards restored on switch) and init. Empty bucket → existing renderEmpty placeholder.
• Card cap stays MAX_ITEMS (50) for DOM; the 100-record bucket remains for P3.2 consumption — consistent with the PR description.

Verification (local worktree, branch head deb4cc9): GUI suite 200/200 pass (79 renderer smoke incl. the 2 new isolation tests), pytest 694/694, doc guard 3/3, node --check clean. CI test run green. Mergeable/clean against master.

No issues found.

[argszero]

✅ LGTM — cycle

Per-session artifact isolation closes a real gap: background-session tool_finished events now go to the sid bucket without polluting the active pane, and switchSession/init restore the pane from the bucket (mirrors the P2.2 tab pattern). Merged master into the branch to resolve the doc-count drift from #665 (195→200) — verified merged result locally: GUI 200/200 ✓ (both P3 file-tree tests and artifact-isolation tests pass), pytest 694 ✓, doc guard 3/3 ✓. CI test run green on the merged head (deb4cc9).

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head deb4cc9 (unchanged since my R732 review): render gate in addToolResult (background-session tool_finished stays in the per-sid bucket, DOM only for current session) + renderArtifacts() restoring each session's cards on switchSession/init. Local: GUI 200/200 ✓ pytest 694/694 ✓ doc guard 3/3 ✓. CI test run green (31462091859). Merge conditions met (3 consecutive ✅ from different cycles).

[argszero]

✅ LGTM — cycle (3rd)

Re-verified against branch head deb4cc9 (includes the master merge syncing docs to 200): per-session artifact isolation — background tool_finished buckets without polluting the active pane, switchSession/init restore from bucket. Local GUI 200/200 ✓ pytest 694 ✓ doc guard 3/3 ✓. CI test run green (31462091859). Merge conditions met (3 consecutive ✅ from different cycles).