Parse SubRip line by line so empty lines are not dropped

[niksedk]

Problem

Subtitle Edit 5 writes a paragraph with no text as N / time / blank / blank (SubRip.ToText). SubRipParser.Parse split the SRT on blank lines and skipped blocks with fewer than 3 lines, so:

• the empty paragraph became a 2-line block and was dropped, and
• the next block started with an empty line, so its time line was at lines[2] instead of lines[1], and it was dropped too.

Plugins that parse and serialize the whole subtitle (RemoveUnicodeCharacters, WordCensor, AmericanToBritish, BritishToAmerican, TypewriterEffect) therefore deleted lines from the user's subtitle, and every SelectedIndices entry below the empty line was shifted. Example: ["Hello.", "", "- Andro!\n- Ben, I've been oozed.", "Last."] parsed to 2 blocks.

Fix

• se5/Plugin-Shared/SubRipParser.cs: Parse is now line based. A block starts at a number line followed by a time code line (at the start or after a blank line); text lines accumulate until the next block start; trailing blank lines are trimmed. The time code pattern also accepts hours past 99 and a leading - for negative times, both of which SE writes and the old pattern dropped. Public API (SrtBlock, Parse, Serialize) is unchanged.
• se5/TypewriterEffect/SubRipParser.cs: same fix in TypewriterEffect's own copy.
• se5/Haxor/Program.cs: Haxor had its own blank-line split. Empty lines lined up by accident, but a line whose text contains a blank line split in two: selecting the line after it changed nothing, and selecting the line itself translated only its first half. It now walks the lines the same way; output is otherwise unchanged.

Verification

A scratch test project referencing SE's libse (SubRip.ToText / LoadSubtitle), run with CRLF and LF:

• The example above: old parser 2 of 4 blocks, new parser 4 of 4. Serialize(Parse(srt)) is byte identical to SE's output, and editing index 2 changes only the dialog.
• Empty first/last lines, consecutive empty lines, a blank line inside a paragraph, number-only text, whitespace-only text, times past 99 hours: both parsers match libse at every index, and libse reads the serialized result back identical. The old parser dropped blocks in 4 of these 6 cases.
• A negative start time stays its own block (the old parser dropped it). Serialize still writes negative times as 0, as before.
• Haxor: the built plugin run for every single-line selection and for all lines; libse reads back exactly the selected lines translated.
• All six plugins build with no warnings.

After merge

Each affected plugin needs a release (workflow dispatch with release), then se5-plugins.json pointed at the new releases:

• american-to-british.yml, british-to-american.yml, remove-unicode-characters.yml, word-censor.yml (Plugin-Shared)
• typewriter.yml, haxor.yml (own changes)

SrtCueDocument on claude/split-dialogs could then use SubRipParser.Parse instead of its own reader.

🤖 Generated with Claude Code