fix: declare unique_constraint on audit_log pkey to swallow Ecto.ConstraintError (OPS-4587)

[palantir-valiot]

Summary

Declare unique_constraint(:id, name: "<table>_pkey") inside changelog_changeset/1 (lib/ecto_trail/ecto_trail.ex:574) so that a colliding primary-key insert on the audit log (the audit_log_pkey / audit_logs_pkey unique constraint) during log_changes/5 (called from update_and_log, insert_and_log, etc.) is turned into a changeset error instead of raising Ecto.ConstraintError. The audit row is best-effort; the user transaction continues to succeed. Added a regression test that forces a sequence collision inside the same tx.

Also bumped version 1.0.3 → 1.0.4, added CHANGELOG entry, ran mix format, upgraded dev/test deps within ranges (benchee, credo + transitives), and mix compile is clean.

Why

Observed in prod (eliot-lamosa-gto-prod):

(Ecto.ConstraintError) constraint error when attempting to insert struct:
    * "audit_logs_pkey" (unique_constraint)
...
(ecto_trail 1.0.3) lib/ecto_trail/ecto_trail.ex:435: EctoTrail.log_changes/5
(ecto_trail 1.0.3) lib/ecto_trail/ecto_trail.ex:315: anonymous fn/4 in EctoTrail.update_and_log/4

Triage (OPS-4587): NOTIFY+FIX, high severity, code_bug in first-party package. The root cause is the absence of a unique_constraint/3 declaration for the dynamically-named audit table's pkey when building the Changelog changeset.

Fixes the exact failure mode described in the stack trace and the Linear issue.

Changes

• lib/ecto_trail/ecto_trail.ex: compute table name from config (default "audit_log"), derive "<table>_pkey", and call Changeset.unique_constraint(:id, name: pkey) in changelog_changeset/1.
• test/unit/ecto_trail_test.exs: new test under update_and_log/3 that:
  1. Performs one successful update_and_log (advances the seq).
  2. Rewinds the *_id_seq via ALTER SEQUENCE ... RESTART WITH.
  3. Performs a second update_and_log; asserts the user op succeeds and only one audit row exists (the colliding insert was swallowed as a best-effort error).
• mix.exs: @version "1.0.4".
• CHANGELOG.md: new ## [1.0.4] entry.
• mix.lock: refreshed benchee 1.5.0→1.5.1, credo 1.7.18→1.7.19 and their transitive deep_merge/statistex (per "keep dependencies fresh" rule).

Test plan

• mix format (clean; any long lines auto-fixed)
• mix compile (clean; only pre-existing redundant clause warning on map_custom_ecto_type)
• mix hex.outdated + mix deps.update benchee credo (upgraded within allowed ranges; lockfile diff reviewed)
• mix test — requires a running Postgres (TestRepo + sandbox + ALTER SEQUENCE in the new test). The agent image has no Postgres, no docker, and no sudo. The regression test is written to go red for "constraint error when attempting to insert" before the fix and green after. CI on the PR will execute it against the real DB.
• Manual review of git diff --stat and full diff vs origin/main (no debug prints, no empty hunks, no scope creep, follows existing patterns, no new comments per "DO NOT ADD ANY COMMENTS").

Closes OPS-4587

[linear-code]

OPS-4587

[acrogenesis]

Closing as a duplicate of #24 — same bug: Ecto.ConstraintError on audit_logs_pkey in EctoTrail.log_changes/5. Filed by the log-agent dedup gap (now fixed in palantir 38438d6). Consolidating on #24.