API manifest: enumerate remaining stdlib modules so unimplemented-API check (#463) covers 100% of supported surface #513

Followup to #463 / #455 DX thread.

The compile-time R005 UnimplementedApi check from #463 only fires for modules whose surface has been enumerated in crates/perry-api-manifest/src/entries.rs::API_MANIFEST. The check in crates/perry-hir/src/lower/expr_member.rs + lower/expr_call.rs is gated on module_has_any_entries(M) — modules with zero entries stay permissive (silent undefined for any property access on them), so the same class of bug Justin hit in #455 can still occur on un-enumerated modules.

Today's coverage: 397 entries across 45 modules (per --print-api-manifest=markdown). Solid but not 100% of node:*.

Concrete gaps to enumerate

Need to audit crates/perry-stdlib/src/lib.rs + crates/perry-codegen/src/.../NATIVE_MODULE_TABLE and reconcile against API_MANIFEST. The drift test crates/perry-codegen/tests/manifest_consistency.rs currently checks one direction (every dispatch row has a manifest entry). It should also check the reverse: any pub mod X; in perry-stdlib without a corresponding manifest entry is a coverage gap that fails CI.

Likely candidates (need verification):

• Modules whose dispatch routes through custom Expr::* HIR variants (per Compile-time error for unimplemented Node / Web APIs #463 the manifest already hand-covers crypto / os / path / process — others may exist).
• New wrappers added under Native package ecosystem v1: open, extensible, with zero developer-facing change #466 Phase 5 — every new perry-ext-* crate should add its module to the manifest as part of the port, but this isn't enforced.
• node: prefix variants (node:fs, node:path, …) — should resolve to the same manifest entries as the bare names. Worth a unit test.
• Web globals at module scope (fetch, URL, URLSearchParams, TextEncoder, …) — these aren't imports, so the property-access path is different. Some flow through Expr::PropertyGet { object: Identifier, ... } rather than NativeModuleRef. Needs separate gating logic.

What "done" looks like

1. Reverse-direction drift test in manifest_consistency.rs enumerates every pub mod in perry-stdlib + every perry-ext-* workspace member's exports + every native-method dispatch table. Any name without a manifest entry fails CI.
2. module_has_any_entries(M) returns true for every supported module, so the permissive fall-through becomes unreachable in practice.
3. A new test in tests/parity/ compiles a TS file that imports each supported module and references a known-bogus symbol, asserting R005 fires.

Why this matters

#463 is shipped and works on the modules it covers, but the framing in the v0.5.585 release ("compile-time error for unimplemented APIs") implies completeness it doesn't have yet. A user on a less-trodden module (async_hooks, dgram, tls.createSecureContext, …) gets the old silent-undefined experience, doesn't know why, and loses the same hours Justin lost.

Refs: #455, #463 (closed), #466 (Phase 2 — adding new modules should auto-enroll them).