fix(scripts): keep a frontmatter field whose value ends in a line terminator

[BryanFRD]

Closes #1207. Follow-up to #1206, from the nit on its review, which merged before this could land on it.

-const FIELD = /^([A-Za-z_]\w*):(.*)$/;
+const FIELD = /^([A-Za-z_]\w*):([\s\S]*)$/;

#1206 claimed its rewrite changed nothing. That held for the alphabet I fuzzed, which had no line terminator in it. . excludes CR, U+2028 and U+2029 while \s includes them, so a value ending in one stopped matching and the field vanished: title:a + CR went from a to "frontmatter needs a non-empty title".

Measured on 400,000 random lines, this time with CR, U+2028 and U+2029 in the alphabet:

	lines without a line terminator	terminator in surrounding whitespace	terminator inside the value
(.*), as merged	identical	1338 fields lost	identical (both drop it)
([\s\S]*), this PR	identical	identical	1718 fields kept that used to be dropped

So [\s\S]* never loses a field the original kept. It differs only where a terminator sits inside the value, and there the original's behaviour was an accident of what . matches rather than a rule: title: a + U+2028 + b has a title, and reporting it as missing was a false failure.

Still linear: without the m flag $ only matches at the end, so the greedy [\s\S]* runs straight there and never backtracks. 0.0 ms at 20,000 characters.

Not reachable through the normal path either way, since the block is split on \r?\n first. node scripts/validate-site-docs.mjs: 169 pages validated.

[ferrfleet]

Correct. Without the m flag $ anchors at end of input only, so ([\s\S]*)$ matches greedily to the end and cannot backtrack, and trim() already strips CR/U+2028/U+2029 from the captured value, so the field comes out clean.

Checked the one case that would worry me, a value that swallows a following key: with CR-only line endings the whole block collapses to a single "line" and title:a\rdescription:b now yields a title of a\rdescription:b where before both fields vanished. It still fails validation, because description never lands in the map, so this cannot turn into a false pass. And it is unreachable regardless since FRONTMATTER requires \r?\n delimiters.

Nit, no action needed here: there is no test harness under scripts/, so the fuzzing evidence lives only in the PR body. If one ever gets added, title:a + CR is the case to pin.