Fix stale and duplicated content in the common.ai provider docs

[kaxil]

The common.ai docs were reorganized recently without rewriting the pages, and a read of every page afterwards found problems in the words rather than the structure. This PR fixes them. No page moves; the sidebar is unchanged.

Correctness. The schema-compare page told readers to set context_strategy="full" to get keys and indexes, but that is the default; the section now shows opting down to "basic" and the parameter bullet, which was garbled, is rewritten. The LangChain connection page called Anthropic OpenAI-compatible; the real criterion is a LangChain model class that accepts api_key and base_url, and the page, the hook page and the hook docstring now say so. The two TypeSafe pages gave different install commands and now agree on the provider's typesafe extra. The LLM operator page lists fallback_conn_ids, which the operator accepts. The document loader example passes a llamaindex connection to the LlamaIndex embedding operator instead of an OpenAI one.

Duplication. Each explanation now has one home and the other pages link to it: the connection-type rename note (troubleshooting), the MCP token_provider and env_provider callables (the toolset page, with the connection page keeping the fields), the LlamaIndex OpenAI-only model check (the hook page), the model resolution order (the connection page, now including the spec-file case that only the hook page mentioned) and the HITL review parameters (the review page).

Stale text. Review-process language ("pre-PR behavior", "tracked as a follow-up", "left out of this change", a "roadmap" section) is gone, as are version-pinned claims ("2.46+", "verified against 2.23.0 and 2.31.1", "legacy flag from 1.x") and a paragraph quoting calibration scores from one classifier release. The one version that stays is the pydantic-ai-slim release that ships the TypeSafe adapter, on the classifier page, because the provider's extra does not pin it (raising that floor is a dependency change, not a docs one).

Model ids. The tree used four OpenAI ids, one of them invented (gpt-5.6-sol, in the quick start and the connection form placeholder). Docs, docstrings, provider.yaml and the example Dags now use gpt-5 with gpt-5-mini as the cheap variant, and claude-sonnet-5 with claude-haiku-4-5-20251001 for Anthropic. Bedrock ids are untouched because they are a different namespace. The per-request batch example keeps gpt-4.1-mini, since it demonstrates a temperature override that the reasoning models reject.

Style. Em dashes are replaced with commas, colons or sentence breaks, tilde H3 underlines match the rest of the tree, extras installs are quoted so the brackets survive zsh, and "PydanticAI" is spelled pydantic-ai when it means the library.

---

• Read the Pull Request Guidelines for more information. Note: commit author/co-author name and email in commits become permanently public when merged.
• For fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
• When adding dependency, check compliance with the ASF 3rd Party License Policy.
• For significant user-facing changes create newsfragment: {pr_number}.significant.rst, in airflow-core/newsfragments. You can add this file in a follow-up commit after the PR is created so you know the PR number.