Add type:playwright visual source for browser-based capture

[jmjava]

Context

We have a working Playwright spike in the course-builder repo (tools/courseforge/demo-capture.py) that captures a full-workflow demo video of the CourseForge A1 LiDAR UI using Playwright's built-in video recording.

Spike results (validated in course-builder)

• Playwright records 1920x1080 WebM at 25fps with smooth mouse automation
• ffmpeg re-muxes to H.264 MP4 (1.5 Mbps, yuv420p) — fully compatible with the existing compose pipeline
• Video + audio mux confirmed: H.264 video + AAC audio combine cleanly via ffmpeg -c:v copy -c:a aac
• The spike script runs headless on Ubuntu and produces ~3.5 MB for an 18-second demo

Proposal

Add type: playwright as a new visual source type alongside the existing type: manim and type: vhs sources.

Integration points

1. config.py — Accept type: playwright in visual_map entries with fields:

   • script: path to a Python capture script (like the spike's demo-capture.py)
   • url: target URL for the browser session
   • viewport: optional {width, height} override (default 1920x1080)

2. compose.py — In compose_segments(), dispatch type: playwright segments to a new PlaywrightRunner that:

   • Launches headless Chromium via Playwright
   • Executes the user-provided script (or a generic "navigate + wait" default)
   • Returns the recorded video path for ffmpeg composition

3. CLI — Add docgen playwright subcommand for standalone capture (useful for testing scripts before full compose)

4. pyproject.toml — Add playwright to project dependencies

Reference

• Spike script: menkelabs/course-builder@feature/playwright-capture-spike — tools/courseforge/demo-capture.py
• Milestone 2 already mentions "headless render — capture slide transitions as MP4 using Playwright or a headless browser"
• Existing pytest-playwright in dev deps confirms Playwright is already in the toolchain

Acceptance criteria

• docgen playwright --script my_script.py --url http://localhost:3300 produces an MP4
• type: playwright segments compose correctly with narration audio via the existing pipeline
• Headless mode works on Ubuntu (CI-compatible)
• Documentation updated with Playwright visual source examples