Treat section titles as cross-reference targets

[bbatsov]

Phase 1 of the Antora/xref work. Real AsciiDoc - and especially the Antora docs I maintain (rubocop/docs, cider/doc) - almost never uses explicit [[id]] anchors; references point at sections by their auto-generated id (e.g. xref:cljs/figwheel.adoc#clojure-cli-setup[] targets === Clojure CLI Setup). Until now adoc-mode only knew about explicit anchors, so none of those resolved or completed.

This generates each section's auto-id the way Asciidoctor does and feeds it into completion (<< / xref:), the xref backend (definitions + completion table), and adoc-goto-ref-label. The id style is detected automatically: a document's own :idprefix: / :idseparator: win; otherwise files inside an Antora component (an antora.yml above them) use Antora's kebab-case style (My Title -> my-title) and everything else uses Asciidoctor's default (_my_title). adoc-section-id-style forces a specific style.

The id-generation algorithm is the fiddly bit, so I validated it against the real asciidoctor CLI over a spread of titles (and a Buttercup spec does the same when asciidoctor is installed). The level-0 doctitle is excluded, and == lines inside code blocks are skipped (reusing the heading validation the nav/imenu code already does).

Next phases (separate PRs): Antora cross-file xref resolution (open xref:page.adoc#frag, complete page targets, project-wide references, include/image resources), then paste-image-from-clipboard.

• The commits are consistent with our contribution guidelines
• You've added tests (if possible) to cover your change(s)
• All tests are passing (eldev test)
• The new code is not generating bytecode or M-x checkdoc warnings
• You've updated the changelog (if adding/changing user-visible functionality)
• You've updated the readme (if adding/changing user-visible functionality)