Verify Windows daemon mutation acceptance and restart persistence against a real graphcoded process

[coneilen]

Summary

Add headless Windows evidence that production DaemonClient mutations reach a real graphcoded.exe, are accepted, and survive daemon restart. This closes the daemon acceptance/persistence residual for five parity rows; all five remain Partial.

Changes

• Add a real-process harness using a unique named pipe and disposable support directory; verify correlated V2 graphChanged replies, persisted graph readback, clean restarts, and the production-derived lifetime mutex.
• Exercise production node create/update/rename, edge create/edit, and sketch-promotion senders; retain edge identity/configuration and node identity across reload.
• Register the Zig helper test with a positive executed-test-count assertion and update only the five owned rows in the current-main parity ledger.

Test plan

RED: Existing unstarted-client queue tests -> they proved command construction and queueing only; with no real-daemon round-trip, daemon acceptance and save/reload could not be asserted.
GREEN: Tools\windows\Tests\DaemonRoundTrip.Live.Tests.ps1 -> REAL_DAEMON_ROUNDTRIP: PASS; correlated replies confirmed mutation acceptance and a fresh daemon process read back persisted identities/configuration.
REGRESSION: Pinned Zig helper filter -> All 1 tests passed; PowerShell parser checks -> both test scripts parsed successfully; git diff --check origin/main...HEAD -> exit 0.

The live harness also observed the production-derived Global\graphcode-daemon-… mutex while each real daemon was running and ERROR_FILE_NOT_FOUND after clean shutdown. It used isolated support state and left the default support tree unchanged. No foreground/UIA, modal, keyboard, app-launch, or native-input evidence is established. The session-ID marker was synthetic, so this verifies mapping-file preservation, not a live backend session. The five ledger rows remain Partial for those missing UI and platform-runtime proofs. The Windows workflow is expected to run because this PR adds a Zig test root and changes Tools/windows/Tests; full make test and make check were not run locally.

Checklist

• I have read the Contributing Guidelines
• I have signed off my commits per the DCO
• Tests pass locally (make test)
• Code follows the existing style (make check)
• I added the test/contract before the implementation and observed the intended behavioral RED failure