Populate llvm.reserved from release LLVM IR bounds-check retention analysis

[Copilot]

bounds_checks_<crate>.json previously emitted llvm.reserved as an empty placeholder. This change makes llvm.reserved reflect whether MIR-discovered bounds checks are still retained after release LLVM lowering by deriving retention from emitted LLVM IR metadata.

• llvm.reserved now carries LLVM-derived retention data

  • Replaced placeholder {} with structured output:
    • analysis: "llvm-ir-release"
    • records: [{ file, line, function, retained }]
  • retained is computed by matching source lines of MIR bounds-check records against LLVM panic callsites.

• Release IR generation integrated into DB construction

  • During bounds DB dump, RAP invokes:
    • cargo rustc --release -- -Cdebuginfo=1 --emit=llvm-ir
  • Wrapper-specific env (RUSTC_WRAPPER, RAP_ARGS) is removed for this subprocess so IR emission is direct and deterministic.

• LLVM IR parsing and source mapping

  • Scans target/release/deps/*.ll (respects CARGO_TARGET_DIR when set).
  • Detects bounds-check panic symbols (panic_bounds_check, slice_index_len_fail).
  • Resolves !dbg !N → !DILocation(line: X, ...) to map retained checks back to source line numbers.
  • Supports multiple !dbg refs per instruction line.

• Failure mode made explicit

  • If IR generation/parsing fails, llvm.reserved is still populated with:
    • analysis
    • error
    • empty records
  • This avoids silent placeholder output and preserves diagnosability.

• Parser-focused unit coverage

  • Added tests for:
    • panic call with debug metadata mapping to retained source line
    • panic call without debug metadata producing no retained line mapping

Example output shape:

{
  "llvm": {
    "reserved": {
      "analysis": "llvm-ir-release",
      "records": [
        {
          "file": "src/lib.rs",
          "line": 42,
          "function": "my_crate::foo",
          "retained": true
        }
      ]
    }
  }
}

---

⚡ Quickly spin up Copilot coding agent tasks from anywhere on your macOS or Windows machine with Raycast.