Caption word-by-word reveal

Karaoke-style captions for reels: instead of a whole phrase popping in at once, each aligned word appears the moment it's spoken, building the phrase up in place; already-spoken words stay lit and the phrase clears when the next phrase starts (same "replace" progression as the static captions).

This is already implemented in the avatar-reel-composer finishing pass (finish_reel.py) — this skill is how to apply/tune it and the invariants to keep. It is the default caption mode (caption_reveal="word").

When to use

• The user wants captions where the words form the sentence as they're spoken ("palabra por palabra formando la frase", karaoke captions, word reveal).
• Applying that look to a new reel, or re-captioning an already-finished reel without regenerating any video.