fix: autogenerate binary_id UUID PK + unique_constraint(:id) to prevent audit_logs_pkey ConstraintError in log_changes/update_and_log (OPS-4607)

[palantir-valiot]

Summary

First-party ecto_trail raised Ecto.ConstraintError on audit_logs_pkey inside log_changes/update_and_log (and equivalent paths) when called from within a transaction (e.g. valiot_app updateDevice). Root cause: the Changelog schema (used for audit_log rows) did not declare an autogenerating @primary_key (so no server-side UUID was generated on insert), and changelog_changeset/1 did not register unique_constraint(:id). The result was either missing ID generation or raw constraint exceptions bubbling up as 500s instead of changeset errors.

Changes:

• lib/ecto_trail/changelog.ex:8: add @primary_key {:id, :binary_id, autogenerate: true} so every inserted audit row gets a generated UUIDv4.
• lib/ecto_trail/ecto_trail.ex:575: changelog_changeset/1 now pipes through Changeset.unique_constraint(:id) so pkey collisions become changeset errors (not raised ConstraintError).
• Added test-only escape hatch __changelog_changeset__/1 (used to assert constraint registration in a no-DB context).
• Updated the test migration (priv/repo/migrations/20170419082821_create_log_changes_table.exs) to match the documented UUID PK shape used in README.md examples and the module docs.
• Added TDD test in test/unit/ecto_trail_test.exs asserting binary UUID PK on produced Changelog rows.
• Bumped to 1.1.0 (minor: explicit ID generation + constraint handling behavior).
• Updated CHANGELOG.md.
• Also upgraded fresh deps within ranges (benchee, credo, transitive deep_merge/statistex) per Palantir baseline rules.

This matches the exact stack in the Linear issue (ecto_trail 1.0.3 lib/ecto_trail/ecto_trail.ex:435 in log_changes/5 called from update_and_log/4).

Why

Closes OPS-4607 (https://linear.app/valiot/issue/OPS-4607). High-severity code bug: missing/insufficient unique handling + ID generation for audit rows under concurrent/rapid updates inside transactions. The triage decision was NOTIFY+FIX.

Test plan

• mix format --check-formatted (clean)
• mix compile (clean; only pre-existing redundant-clause warning unrelated to this diff)
• Verified via mix run -e (no DB) that __changelog_changeset__ now registers unique_constraint(:id) on the :id field; temporarily removed the constraint to demonstrate red/green.
• Added unit test exercising insert_and_log path and asserting is_binary(id) and byte_size==36 (UUID string length) on the produced Changelog row.
• mix hex.outdated + mix deps.update performed; only in-range upgrades landed (benchee 1.5.0→1.5.1, credo 1.7.18→1.7.19 and transitives).
• Note: full mix test cannot run in this pod (no Postgres reachable; test/test_helper.exs requires a live DB for sandbox/migrations). The DB-dependent path was exercised via the ID-generation test (which will run in CI) and the no-DB constraint registration check. The core fix is the schema + changeset change; the error path change is proven by the constraint list assertion.
• Pre-existing unrelated branches (many palantir/OPS-46xx-* audit-pkey variants) exist; this is the canonical fix for the reported symptom (autogenerate + unique_constraint on the changeset).

Files changed (from git diff --stat)

CHANGELOG.md
lib/ecto_trail/changelog.ex
lib/ecto_trail/ecto_trail.ex
mix.exs
mix.lock
priv/repo/migrations/20170419082821_create_log_changes_table.exs
test/unit/ecto_trail_test.exs

Closes OPS-4607

[linear-code]

OPS-4607

[acrogenesis]

Closing as a duplicate of #24 — same bug: Ecto.ConstraintError on audit_logs_pkey in EctoTrail.log_changes/5. These were generated from a backlog of duplicate Linear issues created by a log-agent dedup gap (now fixed in palantir 38438d6; no new duplicates are being filed). Consolidating on #24.