feat(ui): color conversation entries by source in TOC menu

[romusz]

Summary

Update the Table of Contents menu (“Jump to…”, bottom right) with source-aware entries and fix the existing double-highlight behavior. The changes are split into two commits for separate review.

Commit 1: feature — source-aware entries (76152ee5)

• Distinguish human prompts, model replies and progress updates, agent reports, tool results, background-job updates, and notices using existing message metadata.
• Add source-aware icons, accessible names, and light/dark colors. Preserve label, icon, and thumbnail colors during hover and active-position tracking.
• Preserve existing jump targets and entry visibility; no backend or API changes.

Commit 2: fix — hover-only row backgrounds (ba6104d0)

• Fix the persistent row background that appears immediately when opening the TOC and remains alongside the background of a different hovered row.
• Apply the row background only on hover, while preserving the feature's chosen colors and --bg-secondary hover background.
• Keep current-position tracking, automatic centering on open, and jump navigation unchanged.
• Add light/dark regressions covering opening without a persistent background, hovering active and inactive entries, and clearing hover.

Palette preview

The color schemes are just a suggestion, feel free to change them as you see fit.

Validation

• Based on upstream/main at 58877c2f (v0.1407.926103574), with the feature and fix retained as separate commits.
• UI and isolated Go binary builds, TypeScript/Vue type checks, lint, and scoped formatting pass.
• All 85 UI unit-test files pass with UTC and Node's global navigator disabled to match the existing fixtures' assumptions.
• All 26 Chromium checks across toc-categories, popovers, fragment-navigation, tail-first-mount, and compaction-events pass, covering both the source colors and hover-only behavior.
• Unexcluded go test ./server passes with an isolated home directory, private Git configuration, and canonical macOS temporary paths. The initial run hit the existing intermittent TestCompactInPlaceEndToEnd failure; the unchanged full rerun passed.

[romusz]

pls tell me if you'd prefer to separate ba6104d in its own PR