fix(backend): clarify duplicate ID errors

[clonder]

Shows duplicate IDs instead of raw Postgres errors, using one helper for submission and revision.

Closes #7106

PR Checklist

• All necessary documentation has been adapted.
  • no docs changes needed
• The implemented feature is covered by appropriate, automated tests.
  • submission and revision endpoint tests
• Any manual testing that has been done is documented (i.e. what exactly was tested?)
  • ran ./gradlew test ktlintCheck locally, 638 tests passed and 1 skipped

🚀 Preview: Add preview label to enable

[claude]

Claude finished @clonder's task in 3m 28s —— View job

---

Code review

• Read the PR diff and surrounding code
• Check DB constraints the parser relies on
• Check for other consumers of the old error strings
• Review test coverage
• Post inline comments

---

Good change overall. Leaking ${e.cause?.cause} into a user-facing 422 was the actual bug in #7106, and routing everything through one parser fixes submission, revision and the FASTA path together. Two details I specifically checked and can confirm are correct:

• The regex's (upload_id, X) shape holds for all three constraints that can fire here — V1__init.sql:144, V1.16__unique_loculus_accession_for_revisions.sql:3, V1.22__refactor_aux_tables.sql:20.
• Postgres does not quote or escape values in the Key (…)=(…) detail, so the [^,]+ / greedy (.+) split is the right way to handle IDs containing commas and parens. The SAMPLE,C(1) test cases pin that well — nice.
• No website, e2e or docs code matches on the old message strings, so the wording change is safe.

Four inline comments, none blocking:

Where	Point
UploadDatabaseService.kt:117-124	takeUnless drops the stack trace on unparseable unique violations — the one case that needs it. e.message is gone from the log too.
DuplicateRecordId.kt:9-13	Regex depends on English lc_messages and on column ordering; both degrade silently and neither degraded path is tested. A DB-free unit test over the parser would pin it.
DuplicateRecordId.kt:16-21	SQLException.iterator() already walks the cause chain, so the outer generateSequence { it.cause } is redundant.
SubmitModel.kt:340-343	Both call sites discard DuplicateRecordId.field and hardcode the label; the ": $it" suffix is duplicated.

One note on the changed error semantics that is worth a conscious decision rather than a code change: the revision path reports one duplicate (whichever constraint Postgres hits first), where the removed extractDuplicateColumns built a combined message across accession and submission_id. In practice only one constraint fires per statement so this isn't a regression, but a file with both kinds of duplicate now needs two round trips to surface both. The "at least one" wording already sets that expectation, so I'd leave it.

I did not run the test suite — the diff is message-shape only and the assertions are exact-match, so the local ./gradlew test ktlintCheck run reported in the description covers it.
· branch fix/7106-duplicate-upload-id-errors