feat(clips): refine Dictate and empty states

[shawnmcclelland]

Summary

• Refine Dictate history with single-line previews that expand into full transcripts, without duplicate-text flashes or a moving chevron.
• Reveal cleanup, copy, info, and delete controls in an aligned expanded toolbar. Keep the cleanup button recognizable with a green success state after processing.
• Move time and duration into a compact shadcn info popover and remove capture-source badges and transcript tabs.
• Align Library, Spaces, Meetings, and Dictate empty states with shared shadcn composition, clearer copy, responsive wrapping, and one primary action.
• Make Meetings navigation respect its Labs toggle, and use the dedicated Labs server entry point.
• Add localized info-popover labels in all 11 Clips catalogs.

Validation

• 13 targeted Vitest tests passed for Dictate and Library layout.
• Modified-file formatting and diff checks passed.
• Clips typecheck passed; its wrapper still reports existing missing production auth/database configuration in this local environment.
• Lint completed without errors; the existing library-layout effect-cleanup ref warning remains.
• Browser verified single-line truncation, full-text expansion, aligned controls, info-popover interaction, Escape focus return, and outside-click dismissal; no browser errors observed.
• All 11 catalogs contain the new metadata labels. Full localization guards could not run because the local Docs dependency link for remark-mdx is missing.

[github-actions]

Here's a visual recap of what changed:

Open the full interactive recap

[builder-io-integration]

Builder reviewed your changes and found 2 potential issues 🟡

Review Details

Incremental Code Review Summary

The latest head keeps the prior Dictate, empty-state, and localization fixes, but the previously reported owner-only deletion authorization and Library empty/no-results search regressions remain unresolved and are still visible in existing review threads; they are not duplicated here.

Two new Dictate regressions were identified:

• 🟡 MEDIUM — Vocabulary management is hidden until a dictation exists or capture is active, preventing first-time users from configuring recognition terms before their first capture.
• 🟡 MEDIUM — After cleanup creates cleanedText, the expanded card remains on the Original tab because the view effect only handles removal of processed text; the successful cleanup result is not surfaced automatically.

The PR remains high risk because it contains destructive dictation deletion. Focused guards and tests were reported as passing where the repository aliases resolved; some direct test invocations remain environment/alias constrained.

🧪 Browser testing: Will run after this review (PR touches UI code)