fix(#3577): recognize markdown-table phase rows across the roadmap enumeration family (#3599)

* test(#3577): pin table-declared phase resolution across all four surfaces

Failing-first regression for #3577: a GFM table phase listing (Phase header,
id in the first data cell) declared real phases that roadmap.analyze,
roadmap.get-phase, init.phase-op, and the milestone filter all reported as
absent (phase_count: 0 / found: false). Rows pin the lookup, the scope
probe, the analyzer, schema discrimination against the canonical
RoadmapProgress table, fenced-example exclusion, heading+table union
without double-count, decimal ids, and the 999 icebox exclusion.

* fix(#3577): recognize markdown-table phase rows across the enumeration family

A GFM table whose header leads with Phase and whose data rows carry the id
in the first cell is a phase listing — the #2199 bullet blind spot's table
sibling. collectTablePhaseRows (schema-discriminated against the canonical
RoadmapProgress table via matchTableSchema, fence-aware via stripFencedCode,
digit-bearing id shape, 999 icebox excluded) now feeds: the milestone
filter's sole owner scanMilestonePhaseIds, window classification
hasPhaseEntries, both roadmap lookup chains (getRoadmapPhaseInternal +
cmdRoadmapGetPhase, as last-resort tiers after heading and bullet), and
roadmap analyze's enumerator (with the same disk enrichment contract as
headings and a zero-pad-tolerant duplicate guard). init.phase-op resolves
through its existing getRoadmapPhaseInternal fallback.

* fix(#3577): GFM table termination + icebox word boundary in the table scan

Review findings: the row harvest broke only on blank lines, so prose after
a table (a bare date line) could be harvested as a phase id — rows now stop
at the first non-row line per GFM semantics; the 999 icebox exclusion gains
the heading scan's word boundary so 9991 is kept.

* fix(#3577): sanction collectTablePhaseRows in the enumeration drift scanner

The scan's local 999-only exclusion mirrors its parent owner
scanMilestonePhaseIds' deliberate NOT-isSentinelPhaseId choice (a leading 0
is a real decimal phase, #2554), so it cannot route through the sentinel
owner — function-scoped exemption with the documented reason, same entry
shape as the #3262 owner's.

* chore(#3577): add changeset fragment

* chore(#3577): backfill changeset pr number

---------

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

scripts/lint-phase-enumeration-drift.cjs

@@ -119,6 +119,10 @@
  *     read the physical set. Its ENUMERATION path routes through the owner.
  *   - `src/roadmap-parser.cts` `getMilestonePhaseFilter` and its #3262-extracted
  *     set-building owner `scanMilestonePhaseIds` (the same two heading/
+ *     #3577 `collectTablePhaseRows` — the table-scan sibling feeding the same
+ *     membership set; its local 999-only exclusion mirrors the owner's
+ *     deliberate NOT-isSentinelPhaseId choice (a leading 0 is a real decimal
+ *     phase, #2554), so it cannot route through the sentinel owner either).
  *     bullet scans, lifted verbatim so the `roadmap milestone-scope` probe
  *     reads the identical derivation): both deliberately use the local
  *     `999`-only literal, NOT `isSentinelPhaseId`. That canonical predicate
@@ -286,7 +290,7 @@ const FUNCTION_SCOPED_EXEMPTIONS = new Map([
   [path.join('src', 'state.cts'), new Set(['cmdStateValidate', 'cmdStateSync', 'cmdStateRebuild'])],
   [path.join('src', 'roadmap-upgrade.cts'), new Set(['computeMigrationPlan'])],
   [path.join('src', 'smart-entry.cts'), new Set(['detectVerifyFailed'])],
-  [path.join('src', 'roadmap-parser.cts'), new Set(['getMilestonePhaseFilter', 'scanMilestonePhaseIds'])],
+  [path.join('src', 'roadmap-parser.cts'), new Set(['getMilestonePhaseFilter', 'scanMilestonePhaseIds', 'collectTablePhaseRows'])],
   [path.join('src', 'planning-snapshot.cts'), new Set(['buildAllPhaseDirNamesField'])],
 ]);