feat(skills): unify bgm-to-video flows into music-to-video

Replace bgm-to-video, bgm-to-video-new, bgm-to-video-refactor, and the
standalone beat-sync/montage skills with a single music-to-video skill.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

skills/music-to-video/references/motion-primitives/text-spectral-rays/USAGE.md

@@ -0,0 +1,42 @@
+# Using `text-spectral-rays` — it OWNS its wordmark
+
+`text-spectral-rays` is a **self-contained WebGL hero-text renderer**. From ONE rasterized
+glyph mask it draws **both** the solid wordmark **and** the spectral rays that emanate from
+it. Letters and rays share the same mask, so they are always perfectly registered.
+
+## The one trap: never give the word a second source
+
+The ghost / doubled-wordmark artifact comes from splitting the word across two sources:
+
+- ❌ **Wrong** — use the shader as a "rays-only background" and draw the visible letters
+  with a **separate DOM element** (or stack a second text move like `content_swap` /
+  `chromatic_pressure` on the same word). The DOM font (e.g. Inter) and the shader's raster
+  font (Arial Black / Impact fallback) differ in width, shape, and position, so the ray
+  edges never line up with the DOM letters → a misregistered ghost. **Deleting the shader's
+  letter terms does NOT fix it** — the ray mask itself is still the second, misaligned copy
+  of the word.
+
+- ✅ **Right** — let the shader render the wordmark. There is exactly ONE source, so a
+  ghost is structurally impossible.
+
+## Integrate in one pass
+
+1. **It IS the wordmark.** Hide any DOM logo for that word (keep it only as an invisible
+   layout spacer if a tagline/CTA below depends on its box). Never stack a discrete text
+   move on the same word.
+2. **One timeline.** Merge its `progress` / `effectMix` state tweens onto the group's master
+   timeline and repaint via `tl.eventCallback("onUpdate", render)` — no second timeline, no
+   `requestAnimationFrame`.
+3. **Align the cursor to the word.** `mouse.y` must equal the mask's vertical center. If you
+   move the rasterized word off frame-center (e.g. up, to leave room for a tagline), shift
+   the cursor's `y` by the same amount — otherwise the rays cast at the wrong angle.
+4. **Local raster only.** Rasterize the word with a bundled / system font (no CDN font);
+   upload the mask + colour canvases as textures.
+5. **Entrance.** Slam the whole canvas (autoAlpha + a scale punch, `transform-origin` on the
+   word's optical center) on the hit; let `effectMix` bloom the rays just after. The solid
+   letters are present the instant the canvas reveals.
+
+## Pairs with
+
+A background bed (`bg-flow-field`) or **separate** supporting elements (tagline, CTA, rule)
+— never a second treatment of its own word.