Housekeeping: rename the API, add ktfmt, detekt and Konsist

[lemcoder]

Three commits, reviewable in order.

1. Rename the MarkItDown API surface to MikroMarkdown

The factory functions and exception base class still carried the name of the upstream project this was ported from.

• MarkItDown() → MikroMarkdown() (JVM), MarkItDown(context) → MikroMarkdown(context) (Android)
• MarkItDownException → MikroMarkdownException
• CLI command class and its help name now read mikromarkdown
• README updated; the one remaining MarkItDown mention is the upstream project link

Breaking for 0.1.0 consumers, with no deprecated aliases, matching the convert → parse change.

2. Vendor anydoc as a submodule

This is a cherry-pick of #2, which never reached main. #2 was based on document-model-pipeline and merged into it at 22:53 — but that branch had already merged to main at 11:36, so the submodule landed on a dead branch. main's .gitmodules has only markitdown. This commit fixes that.

3. ktfmt, detekt and Konsist

All three run under ./gradlew check, plus a new CI lint job.

tool	version	scope
ktfmt	0.27.0	kotlinlang style, 120 columns, all source sets
detekt	1.23.8	defaults plus config/detekt/detekt.yml
Konsist	0.17.3	ArchitectureTest in library/jvmTest

ktfmt gap: ktfmt-gradle only derives tasks for the common and JVM source sets. androidMain, androidHostTest and androidDeviceTest get none — half the converters live there. library/build.gradle.kts registers matching format/check tasks and wires them into ktfmtCheck and check.

detekt config overrides only rules that clash with deliberate choices here: broad catch in converters (format libraries throw widely and converters degrade instead of propagating), PascalCase factory functions, and complexity thresholds for the format-walking parsers.

Konsist rules, each verified to fail against a planted violation:

1. layers depend in one direction only — model depends on nothing; render and converters depend on model
2. converters do not reach into the renderer — nothing under ..converters.. imports .render.
3. markdown syntax only lives in the renderer — no table pipes, ATX headings, bold markers, image syntax or code fences in string literals outside render/
4. converters implement DocumentConverter and are named accordingly
5. the model stays free of platform dependencies — no java.* or android.* under model/, keeping the model usable from any KMP target

Rule 3 is a text scan rather than an AST rule, so it is blunt on purpose: a legitimate "**" in a converter would trip it. That is the trade that keeps Markdown building out of converters.

Bug found while clearing detekt

The escaping pass tracked line-start position in a mutable flag that was already false by the time it reached an ordered-list delimiter, so both the digit and delimiter branches were dead — 1. item in running text could still open a list. Escaping is now position-based and honours leading indentation.

Other cleanups it prompted:

• PPTX chart extraction went from eight near-identical loops to five toSeries() helpers; bar/bar3D, line/line3D and area/area3D each share a generated series type
• MikroMarkdown.parse selects the converter first, with failure wrapping in a single parseOrFail
• the EPUB manifest predicate and the CLI command class moved into their own named units

Most of commit 3's diff is the one-time ktfmt sweep. Happy to split that into a formatting-only commit if it makes review easier.

Verification

./gradlew check passes: JVM tests, Android host tests, CLI build, detekt, ktfmtCheck and the Konsist rules.

🤖 Generated with Claude Code