test(mcp): gate parallel daemon recovery

[ohdearquant]

Summary

• extract reusable daemon recovery/framing test infrastructure, including an explicit in-process launcher seam for deterministic lifecycle tests
• cover eight concurrent pre-write NoSocket recoverers converging on one usable daemon, while preserving post-write ParseFailure as a terminal exactly-once boundary
• gate the paired recovery scenarios with 25 serial repetitions on Linux and macOS, and align the lifecycle documentation and ADR with the tested contract

Contract details

• production recovery still uses the process launcher and the normal incumbent-exit timeout
• the in-process harness is available only through the explicit fault-injection/test path
• the stable contention oracle is one live, responsive daemon after quiescence; redundant spawn attempts may occur before readiness, as documented
• a post-write parse failure performs no kill, spawn, or follow-up connection
• the default-store sentinel wraps the complete 25-repeat batch, preserving mutation detection without repeatedly hashing the operator store

Validation

• cargo test --manifest-path crates/Cargo.toml --workspace
• cargo check --manifest-path crates/Cargo.toml --workspace
• cargo clippy --manifest-path crates/Cargo.toml --workspace --all-targets -- -D warnings
• cargo fmt --manifest-path crates/Cargo.toml --all -- --check
• RUSTDOCFLAGS=-Dwarnings cargo doc --manifest-path crates/Cargo.toml --workspace --no-deps
• 25 consecutive focused repetitions of cargo test --manifest-path crates/Cargo.toml -p khive-mcp --lib 'daemon::tests::parallel_' -- --test-threads=1
• independent final review: APPROVE, zero findings

Closes #539
Closes #544

AI-assisted contribution: Codex prepared this change and PR description.