docs(seo): PyTorch-Wildlife content depth + long-tail (framework/model-zoo lane)

[rain-Brian]

Summary

Content-depth SEO pass for the PyTorch-Wildlife docs site, claiming the cluster's framework / library / model zoo / training / inference keyword lane. Prose standardizes on "conservation deep learning framework". All API, model, and CLI-style details are sourced from the real package (src/PytorchWildlife/**) and demo/ scripts; nothing was invented.

Stacked on the infra branch: base is seo/pw-infra-structured-data (infra PR #15). On merge of #15, this PR's base retargets to main.

Keyword lane

Priority: pytorch wildlife, wildlife AI framework, conservation deep learning framework, wildlife model zoo, pytorchwildlife pip install. Boundary respected: the framework supports detection / audio / hardware but does not own them. Camera-trap detection links to MegaDetector, audio to MegaDetector-Acoustic, hardware to SPARROW; siblings are framed as the modality owners.

Pages added

• docs/api.md (API Overview): editorial tour of the load-and-run API with runnable detection, classification, detect-then-classify, video, and bioacoustics examples drawn from real method signatures (single_image_detection, batch_image_detection, single_image_classification, process_video, save_detection_*).
• docs/inference-examples.md (Inference Examples): single-image, batch-folder, video, and Gradio web-UI walkthroughs mapped to the actual demo/ scripts.

Pages deepened

• docs/index.md: leads with the value prop, "why a framework not just a model" section, standardized framework phrasing, em-dashes removed, ecosystem block reframed so siblings own their modalities.
• docs/model_zoo.md: H1 promoted to Wildlife Model Zoo; real model tables (MegaDetector V6 variants by license, V5, Deepfaune, HerdNet; classifiers with classes/geography/coverage); related-projects block.
• docs/installation.md: keyword-first title/description for "pytorchwildlife pip install"; tightened sections; next-steps cross-links.

Internal linking

Each new page has 3+ inbound links, descriptive anchors, a hub backlink to microsoft.github.io/Biodiversity, and a "Related Microsoft biodiversity AI projects" block. Nav updated in mkdocs.yml (API Overview, Inference Examples).

Gates passed

• mkdocs build --strict: clean (no link/nav errors; only benign git-timestamp notices for the two new files).
• grep -rn '—' docs/: clean (also fixed one pre-existing em-dash in build_mkdocs.md).
• New pages render; [!TIP] callouts render as admonitions; internal links resolve.
• Front-matter on all 5 edited/new pages: keyword-first title, 110-160 char description, lane tags.
• Originality: paraphrased, no verbatim README copying.
• Touched only docs/** and mkdocs.yml. No src/, pyproject.toml, demo code, or CI changed. site/ not staged.

Claims to verify (@zhmiao)

None. Every model name, class name, version string, license, method signature, and CLI/demo detail was read directly from src/PytorchWildlife/** and demo/. Two minor judgment calls worth a glance:

• No installed CLI exists (no console_scripts in pyproject.toml), so no CLI page was created. Command-line usage is documented via the runnable demo/ scripts on the Inference Examples page instead.
• Species counts (~36 Amazon, ~48 Serengeti, ~44 Deepfaune) carried over from the existing model_zoo.md; Amazon's ~36 is confirmed by num_cls=36 in the source, the others were not independently re-counted from weights.