handle dummy prompts

[dmarx]

particularly, case where first scene no audio

[mattpetters]

mattpetters@991d054

@dmarx I did a quick and dirty solution on this on my fork, just don't write captions if there aren't words.

I did have a question tho, in my testing I found the Whisper API results to be a little unstable / unpredictable (sometimes I got 1 segment, sometimes I got 17, for the same song). Getting something really usable has required turning around and cleaning results or just straight up adding segments manually in the storyboard.yaml

Is that anything we can tune or improve?