feat: serve dig.getManifest locally from the embedded public manifest

[MichaelTaylor3d]

Summary

#176 Phase C: dig.getManifest moves from a blind-passthrough alias to a LOCAL read, resolving the normalized PublicManifest (.dig data-section id 13, Phase A — already merged in digstore's store-libs) embedded in a held capsule's compiled module.

• dig-node-core: new Node::get_manifest + read_public_manifest_blocking (runs on spawn_blocking, mirroring serve_local_blocking). Params {store_id, root} (a capsule identifier — storeId:rootHash, matching dig.getAvailability/dig.fetchRange's shape); no retrieval_key needed since the manifest is PUBLIC, unencrypted data.
  • Module held + manifest present → result = the manifest JSON ({schema_version, entries:[...]}, byte-identical to PublicManifest::to_json).
  • Module held + no manifest section (an older .dig, or a private store) → result: null — never an error (store-format §5.1 backwards-compatibility rule).
  • Capsule not held locally at all → -32004 (same code dig.fetchRange uses on a miss).
  • Malformed store_id/root → -32602 before any filesystem access.
• dig.stage now sets FinalizeOptions.include_public_manifest for PUBLIC stores (matching the digstore CLI's own commit rule) — required because the digstore rev bump below adds this as a new field on that struct.
• Bumped the pinned digstore rev (dig-node-core and dig-wallet, kept in sync per the existing "bump together" contract) to the tip carrying the public-manifest format extension (digstore-core/-store/-compiler/-stage etc., no digstore-guest changes — the vendored guest wasm did not need rebuilding). Added digstore-compiler as a direct dependency of dig-node-core for extract_data_section_blob (a pure wasm-binary-format parse — no wasmtime instantiation needed to read this public section).
• dig-node-service: meta.rs catalogue entry flipped served: "passthrough" → "local" with an updated summary; doc comments in server.rs/rpc.rs updated to match (a clarifying comment was added to rpc.rs explaining Route::Passthrough there is a request-normalization classification, independent of local-vs-relay serving).
• SPEC.md/README.md/USER_JOURNEY.md updated to document the new local behavior (including a new SPEC §5.5.1) instead of describing it as a blind-passthrough alias.

Version bump

dig-node-core 0.1.0 → 0.2.0 and dig-node-service 0.3.0 → 0.4.0 — minor: a new capability (dig.getManifest is now resolved by the node instead of always relaying to the upstream), fully backwards compatible (existing callers of the method still get an answer; everything else is unchanged).

Tests (TDD — written before the implementation)

• dig_get_manifest_returns_embedded_manifest_json_when_present
• dig_get_manifest_returns_null_when_section_absent
• dig_get_manifest_reports_unavailable_when_capsule_not_held
• dig_get_manifest_rejects_malformed_params_without_touching_disk

Both manifest-presence tests build a real compiled .dig module via digstore_stage::stage_and_compile (not a mock) so the data-section extraction + decode is exercised end-to-end. The pre-existing passthrough_alias_methods_are_method_not_found_on_the_node test was updated to drop dig.getManifest from the passthrough-alias list, and the openrpc_drift_guard suite (which dispatches every catalogued method through the real handle_rpc) was re-verified green against the new served: "local" classification.

How verified

• cargo fmt --all -- --check — clean.
• cargo clippy --workspace --all-targets --all-features -- -D warnings — zero warnings.
• cargo test --workspace — all green: dig-node-core 193 passed, dig-node-service 92 passed (incl. the 5 openrpc_drift_guard tests + the 25 server.rs integration tests), dig-wallet 71 passed, dig-runtime 3 passed. Zero failures.
• Confirmed the digstore rev bump resolves to a single coherent version of every digstore-* crate across the whole workspace (dig-wallet had its own independent, now-stale pin — bumped it too to keep the documented "same rev as the node" invariant).

Note: this repo's current required PR checks are "Lint commit messages" (commitlint) and "Check version increment" only — there is no ci.yml running cargo test/clippy/fmt as a required gate yet. I ran all of the above locally per the ecosystem-wide contract; adding a dedicated CI workflow for it would be a separate, repo-wide infra task.