obj_description() can never match: comment OIDs use a different formula than pg_class.oid

[erans]

Found during the final review of #87.

src/metadata/object_resolver.rs computes a table OID with Rust's DefaultHasher. src/ddl/comment_ddl_handler.rs writes that value into __pgsqlite_comments.object_oid, which is exposed as pg_description.objoid.

Since #87, pg_class.oid comes from the canonical 3-character-prefix formula (crate::utils::generate_table_oid), which the SQLite views and the persisted pg_constraint/pg_index/pg_attrdef/pg_depend OIDs all share. The two never agree, so:

SELECT obj_description(c.oid, 'pg_class') FROM pg_class c WHERE relname = 'users';

returns NULL for every table that has a comment. This is what psql \d+ uses to show table comments.

A comment in object_resolver.rs claimed it used "the same algorithm as pg_class view". That was false and has been corrected on the #87 branch; the formula itself was deliberately left alone, because changing it would orphan every already-persisted __pgsqlite_comments row.

Fixing it properly needs a data migration that rewrites __pgsqlite_comments.object_oid to the canonical formula, alongside the code change.

Related: src/functions/catalog_functions.rs has the same problem for the regclass UDF — see the companion issue.