[DIT-13503]: candidate ids hash the line number instead of the occurrence

[laurakoye]

Closes DIT-13503.

Candidate ids hashed (file, line, column, value). The project's identity rule is the opposite: a string is identified by its file, its text, and which repeat it is — never its line number.

Now: id = sha1(file:value:occurrence_index). line and column stay on location, just not in the hash. occurrence_index is already computed before the id loop, so nothing was reordered.

Fixes two things

Ids churned on unrelated edits. An import added above a string shifted its line and changed its id — a delete plus a create to anything keying by id. That's what occurrence_index (DIT-13463) exists to prevent.

Ids could collide. PO plural forms take their location from the same msgid_plural line, so two forms sharing text hashed identically and a consumer keyed by id dropped one:

msgstr[1] "%d rzeczy"   → (file, line, 1, "%d rzeczy")
msgstr[2] "%d rzeczy"   → (file, line, 1, "%d rzeczy")   ← same id

Tests

New src/candidate-id.test.ts — plural forms sharing text get distinct ids; an id survives lines inserted above it. No existing test hardcoded an id.

Breaking

• makeCandidateId is exported: (file, line, column, value) → (file, value, occurrenceIndex)
• every candidate's id value changes

Safe now — nothing persists candidate ids yet, so this is a two-line change today and a migration later.

Also bumps package.json to 0.3.0. With #4, the release note needs three breaks: required snapshot_text, the makeCandidateId signature, and changed id values.