docs(design): close FFB-3/FFB-4/PIP-2 as satisfied by FFB-1 design + PIP-3+5 wire-up

[oferchen]

Summary

Combined closure note for three task chains whose deliverables have
already shipped (or were absorbed into existing design surface).

• PIP-2 (test: add DeleteStats write/read roundtrip tests #2565) - completed retroactively. The design doc that
  PIP-2 was supposed to produce already exists as
  docs/design/parallel-receive-delta-default-on.md, which shipped
  with the parallel-receive-delta scaffold (PR feat(engine): SpillableReorderBuffer wiring + parallel receive delta apply (#1884, #1368) #4319) and was extended
  by the default-on Path B heuristic (PR perf(transfer): enable parallel receive-delta by default via Path B heuristic (PIP-3 + PIP-5) #4666). No fresh design
  artifact required.
• FFB-3 (fix: honour --update and --dry-run flags in daemon receiver #2576) - deferred N-A. The task was conditional on FFB-1
  choosing Option A or B (explicit flush_workers primitive at every
  callsite). FFB-1 (PR docs(design): flush_workers barrier API for ParallelDeltaApplier (FFB-1) #4659) chose Option D - the barrier is baked
  into finish_file. FFB-2 (PR feat(engine): add flush_workers/drain_inflight barrier API to ParallelDeltaApplier (FFB-2) #4665, in CI) ships that bundled
  barrier. Every existing finish_file caller already gets the
  barrier without callsite migration.
• FFB-4 (fix: interop quick-fixes for daemon receiver #2577) - satisfied-by-design. PIP-3 + PIP-5 (PR perf(transfer): enable parallel receive-delta by default via Path B heuristic (PIP-3 + PIP-5) #4666,
  merged) wired ParallelDeltaPipeline into the receiver via
  enable_parallel_receive_delta(). The pipeline's chunk handlers
  reach apply_one_chunk via SlotHandle drop, and the receiver
  finalises files via finish_file. Option D's internal
  flush_workers fires automatically; no explicit call needed at the
  wire-up site.

Followup note in the doc: the FFB-2 implementation required a
spin-then-yield fix around finish_file's Arc::try_unwrap to close a
Windows release-race (tracked in
project_slothandle_decrementguard_release_race.md). Restructuring
DecrementGuard to eliminate the spin would be FFB-5+ if filed.

Doc: docs/design/ffb-3-4-pip-2-closure-2026-05-21.md (145 lines,
markdown only, no source changes).

Test plan

• CI fmt+clippy passes (pure markdown change, no Rust touched).
• Reviewer confirms the three trackers (test: add DeleteStats write/read roundtrip tests #2565, fix: honour --update and --dry-run flags in daemon receiver #2576, fix: interop quick-fixes for daemon receiver #2577) can be
  closed with the cited rationale.