feat(line-captions): add bilingual (English + source language) subtitle rendering

[natashaannn]

User story

As a video editor producing a short in Mandarin, Hindi, or Bahasa Indonesia for a broader audience, I want the burned-in captions to show both the original-language line and its English translation together, so that non-speakers of the source language can still follow along.

As a viewer who speaks the source language but not English, I want the original-language line to remain the visually primary line, so that translation is a supplement, not a replacement.

Background

The line-caption pipeline (docs/implementation-guides/LINE_CAPTION_SHORTS.md) produces one text string per CaptionLine (remotion/types/lineCaptions.ts: { id, speaker, text, startMs, endMs }), and LineCaptionOverlay renders exactly that one string per active line. There is currently no concept of a second language, no translation step anywhere in the pipeline, and no dual-line layout in the renderer.

This issue depends on two others: the source language must transcribe/align/chunk correctly first (multilingual-pipeline issue), and both scripts (source + English, which is always Latin) must render legibly (font-rendering issue only matters for the non-English side here, since English is Latin).

Acceptance criteria

Happy path

Given a completed lines.json in Mandarin Chinese (or Hindi, or Bahasa Indonesia)
When I run the new translation step
Then each CaptionLine gains an English translation string, and the human-editable doc format shows both the original and translated text per line for review/correction.

Given a lines.json with per-line translations populated
When LineCaptionClip renders
Then each active caption window shows both the source-language line and its English translation simultaneously, with the source-language line visually primary (e.g. larger/higher) and the English line secondary (e.g. smaller/below), consistent with the brand's existing typography scale.

Given an English-source video (no translation needed)
When the existing single-language pipeline/renderer runs
Then behavior is completely unchanged — bilingual rendering only activates when a translation is present.

Error path / edge case

Given a line where translation failed or was skipped
When it renders
Then it falls back to showing only the original-language line (no broken layout, no empty second line taking up space).

Given hand-edited translated text in the doc format
When captions:merge runs
Then it updates the translated text field the same way it currently updates text, without disturbing startMs/endMs/speaker.

Out of scope

• Multilingual transcription/alignment/chunking itself (tracked in the multilingual-pipeline issue — this issue assumes a correct source-language lines.json already exists).
• Non-Latin font glyph rendering (tracked in the font-rendering issue — this issue assumes both scripts already render legibly).
• Machine-translation quality/provider evaluation beyond picking a workable option (see Implementation details) — treat translation-quality tuning as a future follow-up.
• Translating in the other direction (English source → other-language subtitles) — this issue is source-language + English only, matching the stated ask.

Technical context

• remotion/types/lineCaptions.ts — CaptionLine needs a new optional field, e.g. translatedText?: string, alongside the existing { id, speaker, text, startMs, endMs }.
• scripts/line-captions/create-line-captions.js — Step 5/5 (chunking, create-line-captions.js:139-151) is where a new translation step would slot in, after chunkIntoLines produces the source-language lines.
• scripts/line-captions/merge-line-captions.js doc-parsing regex (/^\[(\d+)\]\s*(.*)$/ per the implementation guide, Commit 3) needs a compatible extension to also capture a translated-text line per caption id — needs a doc format design that keeps [id] markers parseable (e.g. a second [id-en] line, or a fixed two-line-per-id block).
• remotion/components/LineCaptionOverlay.tsx (per implementation guide Commit 4, prop { lines: CaptionLine[]; brand: Brand }) needs a dual-line layout branch when translatedText is present on the active line.
• Translation approach: no translation API/library is currently used anywhere in this codebase (confirmed no existing translation dependency) — this issue needs to pick one (e.g. a call to an LLM API already used elsewhere in the project, or a dedicated translation API) and add whatever config/env var it requires.

Implementation details

1. Add translatedText?: string to CaptionLine in remotion/types/lineCaptions.ts, and document it in CLAUDE.md's Data Schemas section per the repo convention for new artifact fields.
2. Add a translation step to create-line-captions.js (or a new standalone script, e.g. scripts/line-captions/translate-line-captions.js, run explicitly rather than folded silently into captions:create — keep it a distinct, skippable pipeline stage) that populates translatedText per line via a chosen translation provider.
3. Extend buildLineDoc() (create-line-captions.js:65-80) and the corresponding parser in merge-line-captions.js to round-trip both the original and translated text per line in lines.doc.txt.
4. Add a dual-line rendering branch to LineCaptionOverlay.tsx: when translatedText is present, render both lines stacked with the source line visually primary; when absent, render exactly as today (single line).
5. Add a new npm script (e.g. captions:translate) and document the updated pipeline order (create → translate → merge → render) in docs/implementation-guides/LINE_CAPTION_SHORTS.md.

Additional test scenarios

• Unit test: doc round-trip (write → parse) preserves both original and translated text per line, including for lines with no translation.
• Unit test: merge-line-captions.js leaves translatedText untouched when only the original text is hand-edited, and vice versa.
• Component smoke test: LineCaptionOverlay renders two stacked lines when translatedText is present, and exactly one line when it's absent, at the same active frame.
• Integration test: running the translation step against a small fixture lines.json produces non-empty translatedText for every line.

Hard constraints

• Must not change rendering or data shape for any existing single-language content — translatedText is additive and optional.
• Must not silently drop translation failures — a failed/missing translation must be visibly distinguishable from a deliberate skip (e.g. translatedText: undefined in both cases is fine, but the translation step must log/report failures rather than writing an empty string silently).
• New lines.json field must be declared in remotion/types/lineCaptions.ts and documented in CLAUDE.md, per the repo's Agent Implementation Convention.

Dependency issues

• Depends on the multilingual-pipeline issue (needs a correct source-language lines.json to translate from).
• Depends on the CJK/Devanagari font-rendering issue (needs the source-language script to render legibly alongside English).