fix(fetcher): respect docs page URLs

[chaliy]

What

Fix DocsSiteFetcher so origin-level llms.txt / llms-full.txt is only used for direct index requests or root docs-site URLs.

Specific docs page URLs now fetch the requested page instead of returning the whole-site LLM index.

Why

A page request like https://docs.tenzir.com/reference/operators/where/ was returning the site-wide llms.txt index, dropping the page content the caller requested.

How

• Added a root-docs URL check before probing origin-level llms.txt.
• Kept direct /llms.txt and /llms-full.txt behavior unchanged.
• Added regression tests for root docs index behavior and specific docs page behavior.
• Updated specs/fetchers.md to match the new contract.

Risk

• Low
• Changes only the docs-site fetcher path.
• Main compatibility impact: callers that relied on specific docs pages returning site-wide llms.txt now receive the requested page content.

Checklist

• Unit tests are passed
• Smoke tests are passed
• Documentation/spec is updated
• Specs are up to date and not in conflict
• Security review completed