fix(#3685): report phase-complete write flags from the transaction, not the filesystem (#3826)

* test(#3685): failing-first regression coverage for phase-complete write flags

`phase complete` reports `roadmap_updated`/`state_updated` from
`fs.existsSync(path)`, so both read `true` whenever the file merely exists —
including when the transaction wrote nothing. Add the regression tests that
prove it, plus the negative-space and true-direction pins, before the fix.

New in tests/phase.test.cjs:
- roadmap_updated is false when the transaction rewrites nothing (FAILS today)
- state_updated is false when the transaction rewrites nothing, and stays
  false on a third consecutive run (FAILS today)
- both flags are true when the transaction genuinely rewrites (pins the true
  direction so the fix cannot be tightened into always-false)
- each flag stays false when its file is absent

The STATE.md cases pin the clock via GSD_TEST_MODE + GSD_NOW_MS
(src/clock.cts:43-70) because syncStateFrontmatter stamps a
millisecond-resolution `last_updated:` on every write, which would otherwise
make the no-op unobservable.

Also strengthens four pre-existing `=== true` assertions on these fields that
passed vacuously: each now pairs the flag assertion with a content-changed
assertion against a pre-call snapshot, so the `true` is earned.

Refs #3685

* fix(#3685): report phase-complete write flags from the transaction, not the filesystem

`phase complete` computed `roadmap_updated` and `state_updated` as
`fs.existsSync(path)`, so both read `true` for any project that had the file at
all — including a run that rewrote nothing. The flags are the only signal a
caller has that the rollup landed, so a no-op was indistinguishable from a
successful write and a stale ROADMAP went unnoticed until something downstream
read wrong numbers.

Both flags now reflect whether that file's content actually changed in the
transaction, computed at the existing `writes.push({filePath, before, after})`
sites — the same contract `requirements_updated` has honored since #2316-3, and
the same correction #2640/#2974 already applied to `phase remove`.

Nothing about what gets written changes; only what gets reported.

Fixes #3685

* chore(changeset): backfill PR number for #3685 fragment

---------

Co-authored-by: sim <sim@local>

.changeset/brave-birds-travel.md

@@ -0,0 +1,5 @@
+---
+type: Fixed
+pr: 3826
+---
+**`phase complete` now reports `roadmap_updated` and `state_updated` honestly** — both flags read `fs.existsSync()`, so they were `true` for any project that had the file at all, and a rollup that silently wrote nothing was indistinguishable from one that landed. Each flag now reflects whether that file's content actually changed in the transaction, matching the contract `requirements_updated` already honored. (#3685)