Deduplicate transcripts by youtube_id across users

[howwohmm]

Problem

If 100 users enroll in the same playlist, each video's transcript is fetched independently. No deduplication.

• `worker.py:114-118` — checks `db.get_video_transcript(video_id)` but `video_id` is per-user-course, not per-youtube-video
• Same YouTube video across different users = duplicate fetches, duplicate Groq Whisper costs, duplicate YouTube API hits

Solution

• Add a `transcripts` table keyed by `youtube_id` (not `video_id`)
• Before fetching, check if transcript already exists for this `youtube_id`
• Share transcripts across all users who enroll in playlists containing the same video

Files

• `db.py` — add shared transcript table, lookup by `youtube_id`
• `worker.py` — check shared cache before fetching

Acceptance Criteria

• Transcript fetched once per YouTube video, shared across all users
• Groq Whisper costs not duplicated
• Existing per-video transcript storage still works as fallback