Describe parallel_chunks as chunk planning

## What to build

Align every explanation of `parallel_chunks` with the accepted chunking/fan-out boundary. The setting refines a query plan into more, smaller chunks; it does not execute those chunks concurrently. Fan-out performs execution, and the concurrency setting separately limits how many chunks are in flight.

Audit package documentation, API docstrings, configuration guidance, examples, and architecture prose. Preserve the public name and all behavior. Rewrite only passages that attribute execution or concurrency to the planning dial; leave already-correct division language unchanged.

## Acceptance criteria

- [ ] Every explanation frames `parallel_chunks` as optional chunk-plan refinement or a target chunk count.
- [ ] No explanation says the setting itself performs fan-out or parallel execution.
- [ ] Where performance is discussed, the prose clearly distinguishes more chunks from the separate concurrency cap and fan-out executor.
- [ ] Already-correct chunking descriptions remain intact.
- [ ] The public symbol, signatures, defaults, quota behavior, and execution behavior are unchanged.
- [ ] README and user-guide examples retain their existing calls and rendered behavior.
- [ ] Repository searches find no remaining planning/fan-out conflation around this setting.
- [ ] The documentation build, offline suite, type checker, linter, architecture tests, and dependency contracts pass.

## Blocked by

None — can start immediately.

Start with a repository-wide search for `parallel_chunks`, covering package documentation, API docstrings, configuration guidance, examples, and architecture prose. Update only explanations that conflate chunk planning with fan-out or concurrency, preserving correct division language and all public behavior. Done means the acceptance searches are clean and the documentation build, offline suite, type checker, linter, architecture tests, and dependency contracts pass.

索引モデルが issue の本文から書いたものです。