job_key drops Æ/Ø/Å, so Danish names collapse to unreadable and colliding keys #537

Problem

slugify() folds with NFKD and then encode("ascii", "ignore"). ø and æ have no NFKD decomposition, so they are deleted. å decomposes to a plus a ring, so it becomes a bare a. Danish company and title slugs lose letters and stop being readable, and different words can end up with the same slug: Ørsted becomes rsted, and different words collide: Ærø and Rø both become r.

The docstring at tools/job_key.py:84 and the test at tests/test_job_key.py:153-154 pin this behaviour on purpose ("keyed as rsted since the rule existed"). The concern there is re-keying existing state files. That is a real trade-off, but the folder name derived from the key (/apply, /outcome) is also what users see on disk, and the key is meant to be a pure function of the posting.

Reproduction

tools/job_key.py:66-72 (slugify), reached from make_key at tools/job_key.py:114:

$ python tools/job_key.py --company "Ørsted" --title "Direktør, Åben Æble"
rsted_direktr-aben-ble
$ python tools/job_key.py --company "Ærø" --title "x"
r_x
$ python tools/job_key.py --company "Rø" --title "x"
r_x

Both collision lines were run against the upstream/master version of tools/job_key.py: two different companies get the same key, so the second posting would be treated as already seen.

tests/test_job_key.py:153-154 asserts make_key("Mærsk", "Søfarende") == "mrsk_sfarende".

Suggestion

Transliterate the Danish letters before the ASCII fold, using the standard Danish romanisation: æ→ae, ø→oe, å→aa (and the uppercase forms). Apply it before NFKD, otherwise å is already split.

_TRANSLITERATE = str.maketrans({
    "æ": "ae", "ø": "oe", "å": "aa",
    "Æ": "Ae", "Ø": "Oe", "Å": "Aa",
})

def slugify(text: str) -> str:
    if not text:
        return ""
    text = unicodedata.normalize("NFC", str(text)).translate(_TRANSLITERATE)
    decomposed = unicodedata.normalize("NFKD", text)
    ascii_only = decomposed.encode("ascii", "ignore").decode("ascii")
    return _NON_SLUG.sub("-", ascii_only.lower()).strip("-")

Result: oersted_direktoer-aaben-aeble, maersk_soefarende.

Re-keying: existing seen_jobs.json entries with these letters would no longer match make_key. --audit already reports this as keys_not_matching_current_rule and never renames automatically, so the migration stays a user decision. The PR would update the two pinned test assertions and the docstring, and add cases for å and uppercase Æ/Ø/Å.

Happy to open the PR if this direction is acceptable.