Do not crash on empty WebVTT input or empty SRT cue text

[eeshsaxena]

A couple of the subtitle readers raise on trivial input they otherwise tolerate.

vtt.to_model on an empty document crashes:

import io, ttconv.vtt.reader as vtt
vtt.to_model(io.StringIO(""))   # AttributeError: 'NoneType' object has no attribute 'startswith'

An empty file leaves only the None terminator, and the START state calls line.startswith("WEBVTT") without the None check the other states have. Separately, srt.to_model hits UnboundLocalError on a cue whose text body is empty (a time line followed immediately by a blank line), since subtitle_text is only initialized once a non-empty text line is seen.

I guarded the None line in the WebVTT START state and initialized subtitle_text when the paragraph is created. Added a test for each; the srt and vtt reader suites pass.

[palemieux]

@eeshsaxena Thanks for raising these parsing issues. It looks like the VTT reader never handled VTT files with incorrect signatures according to the specification. I have proposed a fix at:

#536

SRT files will be tackled in a separate PR.

I would very much appreciate your review.

[eeshsaxena]

Thanks for taking this on. Your signature-check approach in #536 is better than my guard, happy to see it handled per the spec. Left one note there about the None return needing handling at the tt.py call site. Fine to close this in favour of #536.

[eeshsaxena]

Closing in favor of #536, which takes a more complete approach to the WebVTT signature handling (proper spec-based first-line check, returning None on a bad signature, and broader test coverage). I left a review there with two small notes. Thanks @palemieux for picking this up.