feat: implement get_schema_snapshot and batch metadata RPCs (#121)

[aesslinger]

What

Implements the three batch metadata RPCs in src/handlers/metadata.rs that were stubs returning "method not found" (#121):

• get_schema_snapshot — what the ER diagram window calls to fetch a schema's tables/columns/foreign keys in one round trip. Opening the ER diagram against this plugin rendered a blank canvas until the host-side fallback in tabularis#822 started composing the same shape from get_tables/get_columns/get_foreign_keys (O(number of tables) round trips instead of one).
• get_all_columns_batch / get_all_foreign_keys_batch — the batch building blocks get_schema_snapshot composes. Not wired to anything on the host yet, but the same shape of gap; shipped alongside for consistency.

Ported the built-in driver's three functions byte-for-byte (tabularis/src-tauri/src/drivers/postgres/mod.rs):

• get_all_columns_batch — batch analog of get_columns: same SELECT (reusing the already-parity-verified row_to_table_column mapper) minus the c.table_name = $2 filter, with c.table_name projected out, ordered by table_name, ordinal_position, grouped by table into the host's HashMap<String, Vec<TableColumn>> wire shape.
• get_all_foreign_keys_batch — batch analog of get_foreign_keys: same SELECT (reusing a new row_to_foreign_key helper extracted from get_foreign_keys' inline mapper) minus the src_cl.relname = $2 filter, ordered by table_name, conname, src_attnum, grouped the same way.
• get_schema_snapshot — composes get_tables + the two batch calls via a pure build_schema_snapshot, returning the host's Vec<TableSchema> wire shape ([{name, columns, foreign_keys}]). Mirrors the builtin's composition (including unwrap_or_default() → empty array for tables with no columns/FKs).

Refactored to avoid duplicating query text: extracted fetch_tables, fetch_all_columns/fetch_all_foreign_keys, row_to_foreign_key, and group_rows_by_table as shared helpers. No rpc.rs changes — all three methods were already dispatched on main, pointing at the stubs.

Wire-shape parity

Verified against the host's RpcDriver deserialization path (tabularis/src-tauri/src/plugins/driver.rs:1456-1551): the three result shapes it deserializes are Vec<TableSchema> and two HashMap<String, Vec<_>>. The TableColumn.is_generated field is #[serde(default)] on the host, so omitting it (as get_columns already does today) is wire-compatible — the host reads the missing field as false, matching the builtin's explicit is_generated: false.

Tests

• Unit (metadata_tests.rs): 4 new build_schema_snapshot tests — zipping tables with their columns/FKs, table-order preservation from get_tables, empty-schema returns an empty array (not null/error), and dropping the table-level comment to match TableSchema's exact shape.
• Live-DB (live_db.rs): 1 new self-contained test exercising all three RPCs against real PostgreSQL 16 (parent w/ PK + enum, child w/ FK + ON DELETE CASCADE, FK-free table), asserting the enum/PK/auto-increment column shapes, the FK ref_table/on_delete + absent-key behavior, and the snapshot's per-table {columns, foreign_keys} arrays.

Verification

• cargo build --release
• Cross-repo 83-test byte-for-byte parity suite (POSTGRES_PLUGIN_BIN against tabularis' src-tauri/tests/postgres_integration parity*): 83/83 pass — exercises the per-row mappers the new batch methods reuse via parity_get_columns/parity_get_foreign_keys/parity_get_tables, so no existing behavior regressed.
• 335 unit + 26 live-DB (1 new) against the local pg-tabularis-test container.
• cargo clippy --all-targets -- -D warnings; cargo fmt --all --check; npx markdownlint CHANGELOG.md — all clean.

Out of scope — filed separately

Hard verification surfaced a pre-existing parity divergence in the shared row_to_table_column mapper: the NULL-default filter is case-sensitive (d == "NULL") where the builtin is case-insensitive (eq_ignore_ascii_case("null")). For a column declared DEFAULT null (lowercase), the plugin emits a spurious default_value: "null" where the builtin omits it. The #121 batch methods inherit this via the shared mapper (consistent with the already-shipping get_columns). The cross-repo parity suite doesn't catch it (no DEFAULT null columns in the seed fixtures). Filed as #122 rather than folded in here for scope discipline; #121 is unaffected and ships the inherited behavior.

Closes #121.

[github-actions]

Version suggestion

Based on this PR's title (feat) and the prerelease:rc label:

Current	1.0.0-rc.4
Suggested next tag	v1.0.0-rc.5

This is informational only — no tag or release is created automatically yet.