Rework amcheck: 3 reports (c1/c2/c3), include system indexes, add GIN

[NikolayS]

Follow-up to #89. Reworks the corruption checks based on feedback.

Changes

3 reports instead of 2 — clearer separation by use case:

Report	Lock	What it checks	When to use
c1	AccessShareLock	btree pages, GIN (PG18+), heap+TOAST (PG14+)	Production primary — safe, fast
c2	ShareLock ⚠️	btree parent-child ordering, sibling pointers, rootdescend, checkunique (PG14+)	Standby — detects glibc/collation corruption
c3	ShareLock ⚠️⚠️	Everything in c2 + heapallindexed + verify_heapam with full TOAST	Standby only — proves every heap tuple is indexed, slowest

System indexes included

All three reports now check pg_catalog and pg_toast indexes too. System catalog corruption is arguably more critical than user table corruption.

GIN support (PG18+)

c1 checks GIN indexes via gin_index_check() on PostgreSQL 18+.

Big fat warnings

c2 and c3 emit prominent warnings about ShareLock blocking writes.

Testing

• PG13: 158 btree indexes checked (no verify_heapam, no GIN)
• PG17: 168 btree indexes + 73 tables checked
• Superuser and non-superuser scenarios verified

[tembo]

On PG10 and earlier this hits bt_index_parent_check(..., heapallindexed := true) (and may fail due to signature/availability). Might be better to explicitly skip c3 on < PG11 with a clear message.

Suggested change

	select current_setting('server_version_num')::int into pg_version;
	select current_setting('server_version_num')::int into pg_version;
	if pg_version < 110000 then
	raise notice '';
	raise notice 'ℹ️ heapallindexed checks require PostgreSQL 11+. Skipped.';
	return;
	end if;

[tembo]

Minor doc nit: this reads like it’s PG18/PG14 only, but the scripts say PG18+ / PG14+.

Suggested change

	| c1 | Quick: btree + GIN (PG18) + heap (PG14) check — safe for production (AccessShareLock) |
	| c1 | Quick: btree + GIN (PG18+) + heap (PG14+) check — safe for production (AccessShareLock) |