Refactor CLI and add preprocessing pipeline

- New CLI structure: -p project, -a action (required flags)
- Add -i import, -f force, -v verbose, --dry-run, --no-cache options
- Add preprocessor.py with chroma key filter (ProRes 4444 output)
- Support background images from shared_assets folder
- Support video metadata JSON files (talkinghead.json)
- Add validation for preprocessed output before render
- Update gnommo.sh with import command and new CLI interface
- Fix Python 3.9 compatibility (Optional[] instead of | None)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

gnommo/transformer.py

@@ -10,7 +10,7 @@ from .models import (
     TimedWord,
     VideoSource,
 )
-from .parser import get_video_duration
+from .parser import get_video_duration, resolve_video_file
 
 
 def build_render_plan(
@@ -26,12 +26,27 @@ def build_render_plan(
     This transforms transcript markers into timed slide events and
     assembles all information needed for the render stage.
     """
-    # For POC: use the first video as the talking head
-    talking_head_id = next(iter(videos.keys()))
-    talking_head = videos[talking_head_id]
+    # Determine talking head source:
+    # 1. If config.talking_head.file is set, use that (may be JSON metadata)
+    # 2. Otherwise, use first video from videos.json
+    if config.talking_head.file:
+        video_path, metadata = resolve_video_file(project_path, config.talking_head.file)
+        # Create a VideoSource from the resolved metadata
+        if metadata:
+            talking_head = VideoSource(
+                file=str(video_path.relative_to(project_path)) if video_path.is_relative_to(project_path) else str(video_path),
+                preprocess=metadata.preprocess,
+                output_file=metadata.output.get("file") if metadata.output else None,
+            )
+        else:
+            talking_head = VideoSource(file=config.talking_head.file)
+    else:
+        # Fall back to first video in videos.json
+        talking_head_id = next(iter(videos.keys()))
+        talking_head = videos[talking_head_id]
+        video_path = project_path / talking_head.file
 
     # Get video duration for end time calculations
-    video_path = project_path / talking_head.file
     total_duration = get_video_duration(video_path)
 
     # Build slide events from transcript markers
@@ -49,6 +64,7 @@ def build_render_plan(
         total_duration=total_duration,
         slides=slides,
         slides_dir=slides_dir,
+        talking_head_path=video_path,
     )