The block boundary gate parses the list marker twice per line

What this is

blockBoundaryReach (src/parse/lines/classify.ts:887) and blockBoundaryOnlyKind (classify.ts:1065) both call parseListMarker on the same line: the first to decide whether the reader lets the marker line win, the second to build the marker arm. The JSDoc at classify.ts:1008-1009 already says the call is made once for both, which is not what the code does. This was the status quo on main before the #232/#261 resolution grouped the arms (reachesLayoutBreak and classifyBlockBody each parsed it too), so it is not a regression, but the coding standard says a fact the caller already holds is not re-derived.

The change

BlockBoundaryReach carries the parsed marker as a payload on the arms that have one (ParsedMarker | undefined, or an arm per case), so the second parse is deleted and the JSDoc becomes true. About 5 code lines and one regex ladder per block-start line. No behaviour change.

Gate

The #182/#242 pins (tests/format/spaced-thematic-break.test.ts), the #232 and #261 pins (tests/parser/include.test.ts, tests/format/include.test.ts) unchanged; bun run test, check, lint green; classify.ts stays under its max-lines ceiling (it is at 1188 lines total; measure). Found by the boundaryresolve re-check and the Sep 7 subtraction map.

Start in src/parse/lines/classify.ts around blockBoundaryReach at line 887, the JSDoc at lines 1008-1009, and blockBoundaryOnlyKind at line 1065. Trace the BlockBoundaryReach arms and their callers, then run the listed pins and bun run test, check, and lint. Done means the marker is parsed once per line, the JSDoc is accurate, behavior is unchanged, and classify.ts remains under its max-lines ceiling.