Add stable tuner and multi-genre chord creator

[pgotta]

What this adds

Stable chromatic tuner

• New Tuner button immediately to the right of Karaoke
• Full-screen/pop-out interface matching Stemmy's existing tool style
• All chromatic notes with sharps/flats display
• Standard, half-step down, Drop D, D standard, Drop C#, Drop C, Open G, and Open D presets
• Selectable microphone/audio interface and adjustable A4 reference
• YIN-style pitch estimation with confidence filtering, rolling median smoothing, cents smoothing, and note-lock hysteresis to reduce bouncing
• Stops Stemmy/Karaoke playback before listening and releases the microphone when closed

Local chord creator

• New Chord Creator button immediately to the right of Tuner
• Genre selection uses checkboxes, so one or several genres can be combined
• Pop, classic rock, alternative rock, post-hardcore, metalcore, punk/pop-punk, indie, blues, folk/country, funk/R&B, jazz, and cinematic/synthwave
• Start from a selected chord, choose 4–8 chords, and generate six options
• No AI and no external API; progression rules and theory logic are local
• Roman numerals, key context, genre character, strumming suggestion, preview playback, chord replacement, transposition, chord diagrams, capo/easy-shape suggestions, copying, and local favorites

Low-risk integration

• The existing app/templates/index.html was intentionally not modified.
• The features live in isolated CSS/JavaScript files and are attached through a small app-factory hook.
• No existing studio, separation, Karaoke, visualizer, MIDI/tab, or backend processing code was changed.

Validation performed

• Python files compile successfully.
• The assembled JavaScript passes node --check.
• Remote Git blob hashes match the locally validated JavaScript/CSS files.
• HTML injection smoke test passed.
• A real browser microphone test and full Windows audio-interface test still need to be performed locally in Stemmy.

Unrelated static-audit findings — deliberately not fixed

These are review notes only, per request:

1. Deep stem-count label mismatch: the UI says “up to 12,” while the Deep preset declares 13 stems and the README also describes up to 13.
2. Karaoke separation failures can lose the useful error: Pipeline.run() can return after a model error, but the Karaoke worker continues into mixdown and may report only “no instrumental produced.”
3. Karaoke retry disk buildup: retrying a failed/interrupted item creates a new project folder without removing the previous failed project folder.
4. Restored Karaoke ZIP edge case: playback heals moved/stale instrumental paths, but the ZIP download route reads the recorded path directly and can omit tracks until another route has healed the stored path.
5. Mixdown length/sample-rate assumption: combined output is sized from the first included stem and later stems are truncated to that length; differing sample rates are not resampled.
6. YouTube stale-file fallback edge case: if the expected yt_<video-id> output is missing, the fallback can select the first matching WAV already in the shared upload directory.

None of the unrelated findings above were changed in this branch.