refactor(frontend): split GeneratePodcastDialog and fix token-count races

[lfnovo]

Summary

GeneratePodcastDialog.tsx had grown to 983 lines and hid two real concurrency bugs. This PR splits the file and fixes the races. Structural moves are behavior-identical; the two behavior changes are called out below.

Structural (behavior-identical)

• ContentSelectionPanel extracted to frontend/src/components/podcasts/ContentSelectionPanel.tsx. It now gets the query client via useQueryClient() instead of a raw queryClient prop, and the handler props are named onNotebookToggle / onSourceModeChange / onNoteToggle.
• Shared selection types/helpers extracted to frontend/src/components/podcasts/generate-podcast-selection.ts (SourceMode, NotebookSelection, formatNumber, hasSelections, getSourceDefaultMode).
• Duplicated sourcesConfig/notesConfig reduction deduplicated: the identical logic in the token-count effect and buildContentFromSelections is now one selectionsToContextConfigs(selections) helper.
• Dead translation cache removed: the tr = {...} block referenced a Proxy-based t() that no longer exists (translations are plain react-i18next now); the panel calls t() directly.

Behavior fixes

• Token/char counter race: the effect awaited chatApi.buildContext serially per notebook on every selections change with no debounce and no cancellation — rapid checkbox toggling produced overlapping request storms with last-write-wins on setTokenCount. Now debounced (400ms) with a generation-counter stale guard: a newer effect run invalidates any in-flight loop, so stale responses can neither accumulate into the totals nor overwrite fresher counts.
• Dialog close timing: handleSubmit closed via setTimeout(..., 500) "to ensure refetch completes" — a sleep racing a network call. useGeneratePodcast's onSuccess already awaits refetchQueries, and mutateAsync only resolves after onSuccess settles, so the dialog now closes immediately after await mutateAsync(...) — deterministically after the refetch.

Verification

Inside frontend/:

• npm run lint — 0 errors (8 pre-existing warnings, none introduced)
• npm run test — 12 files, 85 tests passed
• npm run build — succeeds

[cubic-dev-ai]

No issues found across 4 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{You’re at about 90% of the monthly reviewed-line limit. You may want to disable incremental reviews to conserve quota. Reviews will continue until that limit is exceeded. If you need help avoiding interruptions, please contact contact@cubic.dev.}

_{Re-trigger cubic}