[BUG] Restore visual CI after #2163: five stale toolkit baselines and missing merge gate #2206

Problem

After #2163 switched the Playwright gallery from mocked VS Code toolkit controls to the real toolkit, five baselines were left unchanged. Unrelated PRs and merge-queue runs now fail the same five visual tests.

Missing baselines

• history-rename-editor-{dark,light,high-contrast,high-contrast-light}.png: the real search field places its icon inline, shifting the layout.
• opencode-go-temperature.png: the real checkbox renders inline; expected image is 480×359, actual is 480×340.

#2163 refreshed empty-history baselines but omitted history rename and OpenCode Go temperature. The assertions repeatedly capture stable mismatches, rather than intermittent timing failures.

Evidence and impact

These runs all failed on the same five tests:

• PR #2136: 5 failed, 73 passed. No webview-ui files differ between this PR and its updated base 5cc5a37.
• issue/2084
• feat/l2-chat-diff-default
• PR #2196 merge queue

Expected, actual, and diff images show the toolkit rendering changes. Extension-host visual tests passed.

How #2163 merged

Head 4e740da had an earlier successful visual run, but its final pre-merge visual run failed on these same five tests at 02:56 UTC on October 26. It merged at 03:19 UTC as 5cc5a37.

The current active Protect Main ruleset (15494382) requires check-translations, Ubuntu and Windows platform-unit-test, compile, knip, and e2e-mock. It does not require webview-visual. Thus visual failure does not currently prevent merging. This is a snapshot of current rules, not a historical ruleset audit.

Acceptance criteria

• Regenerate the five missing baselines in the repository's pinned Playwright container and review the image diffs against the intended real-toolkit appearance.
• Confirm the full webview visual suite passes in that container without suppressions or relaxed screenshot tolerances.
• Confirm an unrelated PR / merge-queue run is green with the fix.
• Decide whether webview-visual should be required for merging; update the gate if intended and verify the check is emitted for PR and merge-queue events.

Repair the baselines once on main rather than duplicating them across affected PRs.