[delphyne]

Summary

Two changes that let protoc-gen-markdown and protoc-gen-openapi handle doc-comment Markdown the same way.

Comment cleaning keeps Markdown intact

CommentParser.ParseContext treated any leading non-alphanumeric run as frame chrome, so cleaning ordinary // comments:

• dropped a leading code span's backtick or a reference link's [ (`entity_type` was… → entity_type` was…),
• stripped list markers shared by every line (- a / - b → a / b),
• blanked lines with no alphanumerics, losing code fences and table rules.

Comments with no border lines and no column-zero fence are now treated as plain: only their common leading whitespace is removed and every line in the content range is kept.
Framed, Holub, and JavaDoc comments go through the existing path unchanged.

New module: protoc-utils-markdown

A JVM library (depends on the core library, commonmark-java, slf4j-api) extracted from protoc-gen-markdown's private resolver:

• ReferenceIndex<T> indexes every message, field, enum, enum value, service, and RPC under every name a comment can use, and resolves labels with protoc-gen-markdown's scope rules. A plugin-supplied function maps each element to its own target type.
• ReferenceLinkProcessor<T> is a commonmark LinkProcessor with NONE / WARN / FAIL_ON_INVALID modes, referenceLink=<label>=<URL> overrides, and collected failures with a CodeGeneratorResponse-ready summary.
• MarkdownText renders documents back to CommonMark and flattens them to plain text or a first sentence for fields that can't hold Markdown.

The license allowlist gains BSD-2-Clause (commonmark, as in the plugin repos) and slf4j-api, whose MIT license is declared only in its parent POM.

Release

Consumers pin 1.1.0, so this should be released as 1.1.0 (new module + behavior fix):

• engine-public/protoc-gen-markdown — adopts the shared resolver
• engine-public/protoc-gen-openapi — adopts the shared resolver and cleaned comments
• engine-public/protoc-gen-mermaid — version bump only

Test plan

• ./gradlew build — 7 new CommentTests cases (leading code span, leading reference, shared reference prefix, bullet list, indented list, fenced code, table) plus all existing comment fixtures unchanged; 18 new tests for the markdown module
• Consumed via publishToMavenLocal by all three plugins and the Partner API docs build

🤖 Generated with Claude Code

[andrewtburns]

I did read through each file, and have a good notion of what's going on here.. no obvious errors spotted but kt isn't my finest language