feat(p3-async): cross-memory ptr-pair return for stackful lift

[avrabe]

Summary

Closes the v0.8.0-deferred follow-up under SR-32: stackful async lifts
returning a `string` or `list` across different linear memories.
The v0.8.0 emitter errored out clearly on this path (the
"deferred to follow-up" error); this PR routes it through the same
realloc + memory.copy + nested-indirection + retptr-write sequence the
callback emitter already used.

Refactor

Extracted `emit_ptr_pair_result_writeback` as a shared method on
`FactStyleGenerator`. Both the callback emitter's `is_ptr_len_pair`
branch and the stackful emitter's `is_ptr_pair` branch now call this
single method. SR-12 / SR-17 cross-memory copy contract has a single
source of truth.

Stackful emitter's locals count bumped from 8 to 12:

• 1 step-0.5 scratch
• 4 ptr-pair scratch (src_ptr, src_len, dst_ptr, byte_count)
• 6 nested-indirection patching scratch
• 1 headroom

Tests

• `stackful_ptr_pair_return_emits_realloc_memcopy_retptr_writes` pins
  the wasm shape:
  • `memory.copy` (0xFC 0x0A)

  • =2 `i32.store` (retptr write: new_ptr at offset 0, len at offset 4)

  • =2 `call` (cabi_realloc + lift)

• 208 lib pass (207 prior + 1 new)
• 11 P3 async integration unchanged (callback path untouched in behaviour)

Test plan

• `cargo test -p meld-core --lib` — 208 pass
• `cargo test -p meld-core --test p3_async_lowering` — 11 pass
• `cargo clippy --all-targets -- -D warnings` — clean
• `cargo fmt --check` — clean
• CI green on smithy runners

Refs

• SR-32 — follow-up from v0.8.0
• sub-P3 async lowering: transform stream<T>/future<T>/async exports to RFC #46 host intrinsics #94 — P3 async umbrella
• feat(p3-async): stackful lifting trampoline emitter (#140 phase 2, SR-32) #147 — Phase 2 emitter (where this path was deferred)
• Target: v0.8.1 patch release after this lands

🤖 Generated with Claude Code