Parses a WebVTT file into caption lines.

A WebVTT file is blocks separated by blank lines, and only a cue's text is shown: the lines after its timing line, up to the next blank line. Everything else (the WEBVTT header, NOTE / STYLE / REGION blocks, and a cue identifier, the optional line BEFORE a timing line) is skipped, because none of it follows a timing line within its block.

A cue whose timing cannot be read is dropped whole, and its text with it: attaching that text to the previous cue's timing would show it at the wrong moment.

• start / end in milliseconds; one entry per text line
• skippedCues counts cues dropped for an unreadable timing line, firstSkipped quotes the first of them

The timestamp grammar vttTimestampToMs() accepts: [H:]MM:SS.mmm. The hour field is optional in WebVTT. A comma before the milliseconds is not valid WebVTT, but it has reached the app in production (#584), and rejecting it would drop every cue in such a file rather than show them.

Milliseconds for one WebVTT timestamp, or -1 when it is not one.

converting many stamps, so the regex is compiled once per file rather than per stamp

The timestamp at the start of one side of a timing line: the end side can carry cue settings after it (00:28.876 align:start position:10%), and a CRLF file leaves a trailing carriage return.