feat: natural language playlist generation via MCP onboarding prompt

[gitcoder89431]

Feature

Tuitube already has all the MCP tools needed for an agent to build a playlist from a natural language description — search_tracks, create_playlist, add_to_playlist. The missing piece is a guided prompt template in the tuitube://onboarding resource that teaches Claude to do this end-to-end in one conversation.

Proposed UX

User says to Claude:

*"Build me a 2-hour deep focus playlist from my lofi stations"

Claude:

1. Calls list_stations to find lofi-tagged stations
2. Calls search_tracks with station filter + empty query to get recent tracks
3. Calls create_playlist(name="Deep Focus")
4. Calls add_to_playlist with a curated selection
5. Calls play_track on the first one

All of this is already possible — it just needs documented agent guidance in the onboarding resource.

Implementation

Add a ## Example agent workflows section to the onboarding const in internal/mcpserver/server.go:

## Example agent workflows

### Build a playlist from a vibe description
1. list_stations → identify relevant stations by name
2. search_tracks(station_id, limit: 50) for each relevant station
3. create_playlist(name)
4. add_to_playlist(playlist_id, track_ids) — batch all at once
5. play_track(first track's youtube_id)

No new tools or DB changes needed — purely a documentation/prompt improvement.

Why

Spotify's AI Playlist feature is the most-used AI feature they've shipped. tuitube can offer the same thing, locally, with no ads, for free, via Claude.