[abrichr]

Part of #1674

What changed

• The Postgres code digest in db_snapshot.py now reads rules from pg_rewrite with ev_enabled, so ALTER TABLE ... DISABLE RULE shows as a "(database code)" change. Before, it read pg_rules, which has no enabled state. A view's own _RETURN rule is still left out, the same as pg_rules did.
• The digest now lists unique indexes (pg_index.indisunique, with indisvalid and pg_get_indexdef). A CREATE UNIQUE INDEX that is not a constraint changes which writes the database accepts, and before this it did not show when it was made or dropped.

How it was tested

• Two new live tests in tests/test_db_snapshot_postgres.py: test_disabling_a_rule_shows and test_a_unique_index_made_or_dropped_shows.
• Ran tests/test_db_snapshot_postgres.py and tests/test_frame_database.py against a scratch local Postgres 17 with OPENADAPT_SNAPSHOT_POSTGRES set: 211 passed.
• Control: with main's db_snapshot.py and the new tests, both new tests fail.
• ruff check and format are clean on both files. mypy is clean on db_snapshot.py.

What could not be tested

• Postgres versions other than 17.
• The other items in Whole-database frame on PostgreSQL: follow-ups from the #1673 review #1674 are not in this PR: partitioned table identity, ALTER SEQUENCE INCREMENT in the diff, reading sequences before BEGIN, pooler session pins, citext/hstore, clock defaults, and row text for tables with no key. Table name collision is already refused on main.

[abrichr]

Independent adversarial review: APPROVE

From: Codex. Reviewed actual head 5d7de36 in a fresh detached worktree.

packages/flow/openadapt_flow/runtime/effects/db_snapshot.py:1325 includes each rule's activation mode; line 1336 includes unique-index validity and its complete definition. Both changes are reflected in the database-code diff. No row change was needed to make the probes detect them.

Validation: 211 snapshot/frame tests pass against an isolated PostgreSQL 16.15 cluster. Eight independent live probes pass: all four rule activation modes, expression/covering/NULLS NOT DISTINCT unique-index creation and deletion, and replacing a same-named index's predicate. Every change is missed by the exact parent implementation, detected by this PR, and followed by a stable reread. Full Flow Ruff/format and mypy pass (591 sources).

The environment used locked package dependencies plus pinned psycopg2-binary 2.9.10. Only PostgreSQL 16 was tested in this review. The other items left open in #1674 are outside this approval. Synthetic data, headless resource gate; I did not change or push this PR.