Connect LuaJIT FFI calls to Rust operation handlers

[Ahaeflig]

LuaJIT FFI calls currently stop at the Rust boundary, so callers and impact queries miss native handlers. This adds a source-derived bridge from proven FFI receivers to unique C-ABI exports, then follows literal operations through Lua/Rust wrappers to the handler actually selected and invoked.

Operation edges carry each literal operation to its own handler. A call into an export that dispatches operations is kept as a transport edge, and every walk treats it as a boundary: the export's callers and impact list its Lua call sites, callees and paths end at the export, and code behind the export never fans back out to every Lua caller. The implementation handles lexical aliases and module getters, standard Rust conversions, local Cargo/module identity, function-valued dispatch, and selected declarative macros. It rejects shadowed/mutated receivers, ambiguous exports, guarded or #[cfg] arms and unsupported transforms rather than guessing.

Each file's Lua and Rust analysis is stored in the index under a fingerprint of the analyzers and grammars, so a one-off CLI sync re-analyzes only changed files, and the bridge refreshes only for Lua and Rust files of projects with both languages. Projects without both languages are unaffected, including Lua-only projects' resolution.

Validation against main 6560052 (v1.6.2):

• Full native suite: 6,184 passed, 34 skipped across 473 test files; TypeScript compilation passed. One daemon-cleanup check in mcp-writer-lock.test.ts failed once on a loaded machine and passed three isolated reruns; the bridge does not touch the daemon.
• WASM bridge/provenance/transport coverage: 236 passed across 20 files, including stored analyses, incremental updates, compiled workers, concurrent grammar loading and cross-operation negatives.
• Public RunRustFromLua: 0 → 6 FFI edges. Private gameplay monorepo: 0 → 80 bridge edges, including 72 operation calls; command, history and getter-routed derivation select their respective handlers, with no cross-operation path, callee or caller.
• Node and edge counts remain stable on re-index. All non-timestamp node data and all 266,071 edge rows in the 5,082-file Ruff control match upstream exactly.
• Monorepo fresh index: 55.3 s → 61.2 s (medians of three alternating pairs). Incremental syncs from a new process stay within about 1.5 s of upstream (Lua edit 4.1 s → 5.7 s, Rust 1.8 s → 2.8 s).

Draft: required agent A/B is pending. The Sonnet/high harness attempted two repeats per arm for nine questions, but all 36 attempts hit an account usage-limit/API 429 before tool use. No task timing or adoption results are claimed. A large positive Lua/Rust corpus is also still unverified.

The validation record documents coverage, indexing and sync cost, the pending benchmark, conservative analysis limits and two pre-existing ordinary Rust graph gaps. Existing indexes need re-indexing after upgrading (EXTRACTION_VERSION 28); Cargo manifest edits also require re-indexing. No version, dependencies or native-kernel code changed.

🤖 Generated with Claude Code