Add docgen scene-gen: auto-generate Manim scenes from narration markdown

[jmjava]

Summary

This PR addresses the core design flaw described in issue #1: Manim scene authoring was a fully manual gap in the pipeline. Now docgen scene-gen can automatically generate renderable Manim scenes from narration markdown, eliminating the ??? step.

Problem

The pipeline was: narration → tts → timestamps → ??? → manim → compose

The ??? required manually writing Python Manim scene classes, reading timing.json to place animations at the right timestamps. This was undocumented, time-consuming (~30+ min per segment), and fragile when TTS durations changed.

Solution: docgen scene-gen

New CLI command that parses narration markdown and generates Manim scene code:

docgen scene-gen                     # generate all segments
docgen scene-gen --segment 01        # single segment
docgen scene-gen --dry-run           # preview without writing
docgen scene-gen --force             # overwrite existing scenes

The pipeline now becomes: narration → tts → timestamps → scene-gen → manim → compose

How it works

Narration parsing extracts visual beats from markdown structure:

• # Heading → title card (font_size=36, FadeIn/FadeOut)
• - Item / 1. Item → bullet list (VGroup.arrange, sequential reveals)
• Plain paragraphs → centered body text (font_size=20)
• --- horizontal rules → visual transitions

Timing integration reads animations/timing.json (from docgen timestamps) to distribute beats evenly across the audio duration with buffer between sections.

Generated code follows all lessons from issue #3:

• Text.set_default(font="Liberation Sans") before any Text()
• Uses arrange(DOWN), center(), to_edge() — never absolute coordinates
• Never uses weight=BOLD (Pango font substitution prevention)
• Replaces unsafe unicode (→, ›, —, etc.) with ASCII equivalents
• Minimum font sizes: title=36, bullets=18, body=20
• Dark background (#1e1e2e) with WHITE text

Manual scenes remain supported: generated files are per-segment (scene_01.py) and won't overwrite existing files without --force.

New files

File	Purpose
src/docgen/scene_gen.py	Scene generator module: parsing, timing, code generation
tests/test_scene_gen.py	29 tests covering all components

Changes to existing files

File	Change
src/docgen/cli.py	Added scene-gen command
src/docgen/config.py	Added manim_font property

Acceptance criteria from issue #1

Criteria	Status
docgen init scaffolds projects that produce watchable videos without hand-written Manim	Addressed — scene-gen generates renderable scenes from narration
Regenerating TTS (longer/shorter audio) automatically adjusts visuals	Addressed — scene-gen reads timing.json; re-running after timestamps updates timing
Manual Manim scenes remain supported as an upgrade path	Done — per-segment files, no overwrite without --force
Documentation clearly states the visual authoring workflow	Done — CLI help, docstrings, generated file headers

Testing

• 125 total tests pass (pytest tests/ --ignore=tests/e2e)
• 29 new tests for scene_gen module
• ruff check src/ tests/ — all checks passed

Closes #1