test: replace fixture prose with field placeholders

[jonhoo]

Every test fixture used to carry its own invented Y-Statement — "Store timestamps as UTC", "Retry with exponential backoff", and so on. The parser doesn't care what the words say, so writing that prose was effort that bought nothing, and it actively made snapshot review harder: when in_context and facing_concern each hold a sentence fragment, you have to go back to the fixture to check the parser put each fragment in the slot it belongs in.

Three vocabularies were in play as a result: the README long form (<use case/user story u>) in most of the insta tests, a one-letter form (<u>, <c>, ...) in two unit tests, and hand-written prose in the on-disk fixtures and the YAdr::parse doctest. Fixtures now standardise on a single three-letter form, one per capture slot:

In the context of <ctx>, we faced <con>.

We decided for <opt>, and neglected <alt>.

We did this to achieve <qua>, accepting <dwn>.

We think this is the right trade-off because <why>.

plus <title>, <tail>, and <chg1>/<chg2>/... for change entries. The change descriptions are numbered so that a snapshot would show it if a description were ever paired with the wrong date.

Note that no placeholder may contain a comma or a period: those are what terminate the non-greedy capture groups in YADR_PARAGRAPH_REGEXES. The one fixture that needs bulk, indented_malformed_paragraph, therefore puts its filler inside the angle brackets rather than around them, so that each parameter position stays exactly one placeholder.

Three things deliberately keep their prose:

• Both realistic tests, which parse a Y-Statement taken verbatim from dson. They exist to catch the format only working for prose we wrote ourselves, and they cover what placeholders can't: commas inside chosen_option, a period mid-because, parenthesised asides.
• The format documentation (README, the module docs, the Args doc comment, the YAdr struct docs), which keeps the descriptive long form. Documentation explains what belongs in a slot; a fixture only needs to label it.
• The malformed paragraph in tests/fixtures/broken/malformed.rs, since that sentence is the thing under test.

Some consequences worth knowing about:

• Titles are all <title> now, so tests/cli.rs identifies statements by date rather than by title throughout.
• The YAdr::parse doctest demonstrated that the parser leaves comment wrapping in a field; three-letter placeholders never wrap, so it keeps one deliberately long placeholder that spans the line break.
• Most fixtures still wrap the third paragraph before <dwn> even though it would now fit on one line. That keeps coverage of a capture group whose match spans a line break. The first paragraph of retry.py is wrapped for the same kind of reason — the yadr show reflow assertion needs something to catch — and says so, since unwrapping it would make that assertion pass vacuously.

The 26 affected snapshots were regenerated and reviewed rather than blind-accepted: the diagnostic ones carry width-sensitive caret spans, and each was checked to still underline the paragraph it is meant to. cargo test --all-targets, cargo test --doc, clippy, and the dogfood yadr check over this repo all pass.