Remove the build manifest, execution plan, and enforcement-edits policy (#673)

The project stopped following the build manifest some time ago, and the
execution plan and enforcement-edits policy were written for the same
rebuild. This removes all three, the two skills that drove the manifest,
and every reference to them.

## Removed
- `docs/architecture/build-manifest.md` and the `do-next` and
`review-slice` skills
- `docs/architecture/0002-execution-plan.md`
- `docs/architecture/enforcement-edits.md`

## Behavior changes in tests
- `SymbolCoverageGridTest` and `FeatureMatrixTest` no longer read the
manifest. A not-applicable cell's blocker must now name an issue or an
`RFC 1 §` section; manifest step ids and `Plan 0002 §` sections are no
longer accepted. No registered blocker used either.
- `testABlockerNamingNoSliceIsReported` is renamed
`testABlockerNamingNoIssueOrSectionIsReported` in both. No test is
removed.

## Reference cleanup
- Step and plan citations are stripped from comments and assertion
messages across `src/` and `tests/`. A few comments that described
finished migrations as still to come now describe the current code.
- Pointers to `enforcement-edits.md` are removed from `CLAUDE.md`,
`phpstan.neon`, `deptrac.yaml`, `bin/check-baseline-shrink`, the rule
classes under `tests/Architecture/`, and `.claude/settings.json`. The
rule in `CLAUDE.md` against editing rules, allowlists, and baselines is
unchanged.
- `tests/Parity/README.md` drops the plan framing and names the classes
that exist today.

## Also
- `.deptrac.cache` is gitignored.

## Not in this PR
- `phpstan.neon` rule messages still end in "(build-manifest step-N)",
and the header comment in `bin/check-baseline-shrink` still names a
manifest step. Changing the messages requires regenerating the baseline.
- `docs/architecture/0001-foundational-architecture.md` remains,
including its two mentions of the manifest.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Author: Firehed

.claude/settings.json

@@ -19,8 +19,7 @@
       "Edit(tests/Architecture/**)",
       "Edit(bin/**)",
       "Edit(.github/**)",
-      "Edit(.claude/**)",
-      "Edit(docs/architecture/enforcement-edits.md)"
+      "Edit(.claude/**)"
     ]
   }
 }

.claude/skills/do-next/SKILL.md

@@ -2,4 +2,5 @@
 /tests/Fixtures/vendor/
 /composer.lock
 /.phpunit.cache/
+/.deptrac.cache
 /phpcs.xml

.claude/skills/review-slice/SKILL.md

@@ -129,7 +129,6 @@ Some of them cement the old shape.
 When one fires against a change that follows the design above, stop and ask the human.
 Do not route around the rule, and do not bend the design to satisfy it.
 Never edit a rule, an allowlist, a baseline, or `bin/check-baseline-shrink` yourself.
-`docs/architecture/enforcement-edits.md` classifies every such edit.
 
 When considering edits (or avoiding them), remember: the project and architecture goals are more important.
 

.gitignore

@@ -43,7 +43,7 @@ foreach ($files as $file => $total) {
         fwrite(STDERR, sprintf(
             "%s grew (%d -> %d): baselines only shrink. Route the new" .
             " violation through its authority. Growth is permitted only" .
-            " for a newly added check; see docs/architecture/enforcement-edits.md.\n",
+            " for a newly added check.\n",
             $file,
             $base,
             $head,

CLAUDE.md

@@ -3,7 +3,7 @@ deptrac:
     - src
   # Layer-dependency contract (default-deny): an edge not listed in the ruleset
   # fails analysis. Adding a layer tightens; adding a ruleset edge loosens and is
-  # the human's edit only. See docs/architecture/enforcement-edits.md.
+  # the human's edit only.
   layers:
     - name: Root
       collectors: