Convert the 3.5.x site documentation from APT to Markdown

[slachiewicz]

The 3.5.x line still carried 37 APT pages; master converted its own in #3422 plus the two follow-up
fixes. This brings the branch across.

Eighteen pages are taken verbatim from master — their APT sources are byte-identical to master's
pre-conversion tree, so master's reviewed Markdown is the correct output for them, including the table
and quote fixes from #3424 and #3425. The other nineteen had drifted: 3.5.x still documents
surefire-junit3/4/47/testng provider selection, the POJO provider, the JUnit3 security manager,
TestNG suite XML files and testnames, none of which exist on master. Those pages were seeded from
master and the 3.5.x wording ported back in, rather than re-run through the converter, so they inherit
master's heading, fence and conditional handling.

Two pages exist only here (examples/pojo-test, featurematrix) and went through doxia-converter.
api, developing, pojo-test and featurematrix contain no Velocity at all, so they lose the
.vm suffix, matching what master did for the first two.

Renames are a separate commit — please merge or rebase, not squash, or git log --follow stops at
the rename.

The comparison found four things a green build would not have:

• skipping-tests has a #{if}/#{else}/#{end} spanning most of the file, and the skipITs paragraph
  lives in the #{else} branch, so it is failsafe-only. Porting master's text unguarded leaked it onto
  the surefire page.
• APT does not interpret <<<x>>> inside a +---+ block, so the provider-selection pseudo-code
  rendered that markup literally; it is kept literal.
• featurematrix lost table-bordered, and is now a raw HTML table.
• the heading Running 'testnames' in test tag had its apostrophes turned typographic, changing the
  anchor from .27 to .E2.80.98. Escaped as ', the same fix as Restore the straight quotes the FAQ conversion turned typographic #3425.

One difference is deliberate: history.md.vm stays a pipe table and so still loses table-bordered.
Writing it as raw HTML would restore the border but strip class="externalLink" from every link in the
table, which is the worse trade. master made the same call.

Verified: mvn site before and after on maven-surefire-plugin, maven-surefire-report-plugin,
surefire-api (109 pages) and on maven-failsafe-plugin, which re-renders these pages from the
site-source artifact (54 pages). Compared on visible text plus metadata, on block structure, and on
heading anchors. No page missing, no anchor changed; the only remaining diffs are history.html's
border and the generated surefire.html / dependencies.html reports, which differ on test duration
and dependency size between runs.

This change was created with AI assistance.

[slachiewicz]

Force-pushed one fix: api, developing, examples/pojo-test and featurematrix drop the .vm
suffix, and the first commit was renaming them to *.md.vm while the second renamed them again to
*.md on top of the content rewrite. The combined change fell below rename-detection similarity, so
those four showed as add/delete and git log --follow stopped at the conversion commit — defeating the
reason for splitting the commits at all. git log --follow on them reached 1 commit before, and 8, 24,
8 and 14 respectively now.

The rename commit now renames each file straight to its final name, and the second commit touches
content only — 37 files, all M, no add/delete pairs. The resulting tree is byte-identical to what was
pushed before, so the site comparison in the description still stands unchanged.

Also worth calling out, since it is a deliberate deviation from master rather than part of the 3.5.x
delta: surefire-api/src/site/markdown/index.md drops the <th>Term</th><th>Definition</th> header
row that master's conversion introduced. The APT table here has no header row, so master invented one;
removing it keeps this page rendering what the 3.5.x site rendered.

This comment was created with AI assistance.