blockmap: a lazy-continuation admonition keeps its marker and its indent

[karlkfi]

Follow-up to #19, which only covered half the construct. This is the half that was silently destroying callouts.

Problem

MkDocs and Python-Markdown accept both of these:

!!! tip "Blank line after the marker"

    Body indented four spaces.

!!! tip "No blank line"
    Body indented four spaces.

The first is a paragraph plus an indented code block, which #19 recognizes. The second is a single paragraph to CommonMark, because the indented body is a lazy continuation rather than a code block, so it never reaches that recognition and reflows as ordinary prose:

$ printf '!!! tip "Title here"\n    The recommended shape is the v2 API at\n    a decomposed resource. Second sentence here.\n' | mdreflow -
!!! tip "Title here" The recommended shape is the v2 API at a decomposed resource.
Second sentence here.

MkDocs then renders that as a paragraph beginning with a literal !!! tip. The callout is gone.

Why the marker alone is not enough

I tried the obvious fix first, a lineBoundary dialect rule like the GitHub alert marker. The marker survives, but the body loses its indent:

!!! tip "Title here"
The recommended shape is the v2 API at a decomposed resource.
Second sentence here.

Python-Markdown needs the 4-space indent, so that is still not a callout. The alert case works only because > is the paragraph's container prefix and gets reproduced; an admonition's indent is not a container prefix.

Proposal

build already handles exactly this shape for a footnote definition: a marker first line whose body must stay indented, done by appending to ContPrefix. An admonition needs the same plus an immovable marker line.

if admonitionMarkerRE.MatchString(trimmed[0]) {
    boundary[0] = true
    contPrefix += "    "
}

Result:

!!! tip "Title here"
    The recommended shape is the v2 API at a decomposed resource.
    Second sentence here.

Marker preserved, body reflowed, indent intact. No skip, and the body still gets sentence-per-line.

Verification

• go test ./... green, plus a table case asserting both the boundary and the prefix.
• Constructed edges: the blank-line spelling still routes through blockmap: reflow MkDocs admonition bodies under an opt-in dialect #19, prose merely opening with !!! is untouched (the marker pattern requires a type word), a marker with no title works, and one nested in a list item keeps its container prefix.
• Full mkdocs build over the 263-file docset this was found on, before and after: structural page differences went from 7 to 0.

The two pages that still differ structurally are the reflow fixing latent bugs in that source, not regressions: a sentence wrapped so 10. landed at line start had been rendering as a stray <li>, and a heading that was missing from the page nav now appears.

Unrelated pre-existing failure, for your awareness

Fuzzing this branch surfaced an idempotency failure on "*\n\t**\n**\n:::\n0". It reproduces on unmodified main with the same seed, so it is not from this change and I have not tried to fix it. Happy to file it separately if it is not already known.

[jbeda]

Merged (rebased onto the fresh v0.1.5 — your CHANGELOG entry moved to the new Unreleased section with a #31 reference). Took it ungated after weighing the dialect question: boundary-protection for dialect markers is the always-on posture here (GitHub alerts, Docusaurus fences, Hugo shortcodes all get it regardless of dialect), and this is render-preserving under goldmark, so it belongs in that family rather than behind --dialect mkdocs, which stays reserved for render-changing recognition. The 7→0 mkdocs-build verification is the kind of evidence that makes these easy to take. Your idempotency find is filed as #32 with credit — confirmed pre-existing, confirmed backstop-absorbed for users, and flagged as the second ::: -neighborhood find in a day. It's first in line for v0.1.6, which at the current rate you're setting will be soon.