#20 Code page conventions (breadcrumbs, anchors, clone block)

Context: Owner review round — code page must feel like GitHub/Gitea/GitLab for file view, download, clone. Tasks (keep tiny, no new deps):

1. File view layout: breadcrumb path (each segment linked), file meta bar (name, size humanized, lines count, raw button, copy-path affordance as plain text? no JS — show full path text), line numbers clickable anchors (#L12) so links are shareable.
2. Directory view: breadcrumb + listing with sizes; parent .. row.
3. Downloading: per-file raw download (exists) + directory tarball? (dir → generated .tar.gz on the fly? NO — build-time: generate per-directory tarballs? too fat. Decision: single repo.tar.gz + repo.bundle + raw files only, but SURFACE them properly: prominent Clone/Download block on code tab with tabs? no JS — stacked sections: Clone (clone cmd + bundle + tarball) and per-file Raw links. Add ? no. Keep: clone block with copyable commands as selectable <pre>, bundle/tarball with sizes+shas.)
4. Cloning docs on code tab: git clone <url> (if configured) + git clone repo.bundle + tarball extract cmd. If no clone URL, bundle-first presentation.
5. Tests: breadcrumb anchors, #L anchors, size/line meta asserts, clone block asserts (with and without clone-url). Success rubric:

• Breadcrumb + #L anchors + meta bar on file pages; clone block prominent with all three methods; no-JS throughout. Swarm: Owl. Branch m6/code-conventions.