docs(models): remove traditional ML (ONNX) inference (spiceai/spiceai#11684)

[claudespice]

Summary

spiceai/spiceai#11684 deleted the model_components crate and all traditional machine learning (ONNX/Tract) inference, including the /v1/predict and /v1/models/{name}/predict HTTP endpoints. The spiced build no longer carries any ONNX dependency, and the ModelType (ML vs LLM) distinction is gone — models are LLM-only.

Verified against today's origin/trunk (not just the merged diff): crates/model_components is gone, crates/runtime/src/http/v1/inference.rs is gone, no predict route remains in http/routes.rs, and no ONNX dependency exists in Cargo.toml.

This PR removes the now-inaccurate ONNX/traditional-ML documentation from the Models component and reference pages:

• features/machine-learning-models/index.md — converted to a deprecation notice (page kept to avoid broken links; links to the v2.1 docs, which correctly document the feature for that release).
• components/models/index.md — dropped the now-empty ML Format(s) column; intro and local-model example updated to LLM-only.
• components/models/filesystem/{index,deployment}.md — removed the ONNX format clause, the ONNX example, the ONNX row in the supported-formats table, and the ONNX build-flavor mention.
• components/models/huggingface/{index,deployment}.md — removed the ONNX "predict taxi trips" example, the "ML models only support ONNX" limitation, and the ONNX quantized-format mention.
• reference/spicepod/models.md — LLM-only intro; the datasets field now correctly documents its current behavior — an LLM tool table allowlist (create_table_allowlist / with_table_patterns in crates/runtime/src/model/nsql.rs), not ML-inference dataset preselection.
• getting-started/spicepods.md — "language or traditional ML models" → "language models".

Versioned docs

vNext-only. v2.1.x (the latest cut release) still ships ONNX and correctly documents it, so no versioned snapshots are edited.

Out of scope (follow-ups)

• components/models/spiceai.md (Spice Cloud Platform provider) still shows ML-inference examples and needs an LLM-usage rewrite — deferred pending confirmation of the current spice.ai LLM from: form.
• The OpenAPI-generated /v1/predict API pages (api/HTTP/get-model-predict.api.mdx, post-batch-predict.api.mdx) require regenerating public/openapi.json via the OpenAPI generator — a separate generator-driven change.
• The embeddings pages' ONNX format entries (unrelated to model_components) and vision-prose mentions of ML predictions are better handled by a docs-accuracy review.

Source PRs

• remove(ml): delete model_components crate and ONNX/Tract ML inference spiceai#11684 — remove(ml): delete model_components crate and ONNX/Tract ML inference

Test plan

• cd website && npm run build passes (Docusaurus onBrokenLinks: 'throw'; fixed one relative-link resolution)
• Versioned-docs propagation checked — removal is vNext-only (v2.1.x correctly retains ONNX)
• Files updated: 8

[github-actions]

✅ Pull with Spice Passed

Passing checks:

• ✅ Title meets minimum length requirement (10 characters)
• ✅ Has at least one of the required labels: area/blog, area/docs, area/cookbook, dependencies
• ✅ No banned labels detected
• ✅ Has at least one assignee: claudespice

[github-actions]

🔍 Pull with Spice Failed

Passing checks:

• ✅ Title meets minimum length requirement (10 characters)
• ✅ Has at least one of the required labels: area/blog, area/docs, area/cookbook, dependencies
• ✅ No banned labels detected

Failed checks:

• ❌ At least one assignee is required for this pull request.

Please address these issues and update your pull request.

[github-actions]

🚀 deployed to https://d4afcd3e.spiceai-org-website.pages.dev