fix(docs): correct structured-data defects in JSON-LD output

[AlexBodner]

Five independent defects in the JSON-LD emitted by docs/hooks/schema_inject.py and docs/overrides/main.html. Grouped because they share the same two files and the same reviewer context; none is coupled to the others.

Found while auditing the docs for search/AI-answer visibility. Unlike #582 and #583, none of these is fixed by cutting a release — they are live on /develop/ today.

1. FAQPage markup contradicts the visible answer (and the benchmark tables)

The schema still claimed:

Start with ByteTrack — it performs best across two out of four benchmarks

The visible answer in docs/index.md was already corrected to credit McByte for accuracy and BoT-SORT for camera motion, and the Default HOTA tables agree with the visible text, not the markup. Google requires FAQPage markup to match visible content, so this was both a factual and a policy problem. Synced, with a BENCH-XREF note so the two are updated together per docs/AGENTS.md.

2. BreadcrumbList was never emitted — anywhere

Verified zero BreadcrumbList blocks on /latest/ and /develop/, despite _build_breadcrumbs() being fully implemented.

Cause: _find_in_nav only appended a section when it resolved to a URL. But nav sections (Home, Usage, Tuning, Trackers) are pure groupings with no page behind them, so mkdocs gives them no URL and no file — section_path came back empty for every page and the if not section_path: return None guard bailed out every time.

Now records sections by name (position + name are sufficient per schema.org) and drops the guard.

Note that intermediate ListItems for URL-less sections carry name but no item, because there is no page to link to. That is valid schema.org, but Google's rich-result guidance is stricter about intermediate items, so display is not guaranteed. The fallback is the status quo, no breadcrumb shown, so this cannot regress anything.

3. Organization.logo was relative and inconsistent

{{ 'assets/...' | url }} resolves relative to the current page, so the same @id (roboflow.com/#organization) emitted assets/logo-trackers-violet.svg on the homepage and ../../assets/logo-trackers-violet.svg two levels down — never an absolute URL. Now derived from site_url.

4. SoftwareApplication missing softwareVersion and license

Both are standard properties and both were absent. Version is read from the installed package through a new on_config hook so it cannot drift from pyproject.toml; license is the Apache-2.0 URL.

5. TechArticle.image was the brand SVG on every page

SVG is not an eligible format for Article image rich results, and one generic logo represents no page's content. publisher.logo already carries the brand mark, so the field was redundant as well as ineligible. Removed.

Verification

mkdocs build on this branch, inspecting the generated HTML:

Check	Result
All JSON-LD blocks parse as valid JSON	✅ 4 / 3 / 3 blocks on homepage / ByteTrack / Install
Breadcrumbs emitted	✅ Home > Trackers > ByteTrack, Home > Usage > Install Trackers
Organization.logo absolute + identical across pages	✅
Logo path resolves under the versioned prefix	✅ /latest/assets/logo-trackers-violet.svg → 200
softwareVersion / license	✅ 2.6.0 / Apache-2.0 URL
TechArticle.image	✅ absent
FAQ markup matches the rendered answer	✅

ruff check and ruff format clean. No Python source outside docs/hooks/ touched.

🤖 Generated with Claude Code