DWARF Phase 3: synthetic DIEs for adapter / inlined code (#130)

[avrabe]

Sub-issue of #130 — Phase 3 (final phase) of the witness-mapping epic.

Phase 2 (#143) gets source attribution working for original component code in the fused output. Phase 3 covers the generated code: cross-component adapters that meld itself emits and that have no original source.

Scope

Per #130's plan §"Phase 3 — adapters and inlined code":

• Synthesise DIEs that point at a placeholder source file `` line N, where N encodes the adapter's role:
  • 1 = memory.copy / canonical-ABI transcode loop
  • 2 = `cabi_realloc` trampoline
  • 3 = canonical-ABI lift
  • 4 = canonical-ABI lower
  • (extend as new adapter classes ship)
• Witness's truth-table view then surfaces these as "adapter" branches that don't require source-level MC/DC coverage, rather than as un-attributed gaps.
• Alternative explicitly considered: accept the gap. Witness already has a strict-per-`br_if` fallback when DWARF is absent or sparse. Decision-time question.

Done criteria

• Synthesised DIEs covering every adapter-class function body
• Witness verifies adapter ranges resolve to `:N` rather than `unknown`
• LS-N for "adapter branches accounted in MC/DC line map"

Dependencies

• Blocked by DWARF Phase 2: address remap (#130) #143 (Phase 2). Phase 2 establishes the rewrite/append pipeline this hooks into.

Out of scope

• Variable-level debug info — out per DWARF / source-map remapping for fused output (witness MC/DC integration) #130.

Rivet artifact

• New requirement SR-36 — "Synthesised DWARF DIEs for meld-generated adapter functions".

Milestone

v0.11.0