Refresh statistics parser settings between query files - #24738

Which issue does this PR close?

• Closes Refresh SQL parser settings between statistics query files #24719.

Rationale for this change

The statistics runner previously captured SQL parser settings only once. A successful SET in one query file therefore changed the session but did not affect parsing of later files, which made ordered query suites behave inconsistently with session state.

What changes are included in this PR?

• Read the current session's parser options immediately before parsing each query file.
• Extract file processing into a focused helper without changing the existing same-file pre-parsing behavior.
• Document that parser-setting changes still do not affect later statements in the same already-parsed file.
• Add ordered-file regressions for MySQL dialect propagation and parser recursion-limit propagation.

Are these changes tested?

• cargo test -p datafusion-benchmarks --lib statistics::tests -- --nocapture (7 passed)
• cargo test -p datafusion-benchmarks --lib 'statistics::tests::refreshes_' -- --nocapture after rebasing (2 passed)
• cargo check -p datafusion-benchmarks --lib
• cargo clippy -p datafusion-benchmarks --lib -- -D warnings
• cargo fmt --all -- --check
• Ablation: both new regressions fail when parsing uses stale default settings.

The broader cargo clippy -p datafusion-benchmarks --all-targets --all-features -- -D warnings could not complete locally because the optional snmalloc target requires cmake, which is unavailable in this environment. It produced no Rust diagnostics before that build-tool failure.

Are there any user-facing changes?

Yes. Parser settings changed by an earlier statistics query file now apply when later query files are parsed. There are no public API changes.

This draft was prepared with AI assistance and has not yet received human review. I understand the implementation end-to-end; there are no known design assumptions beyond the documented same-file parsing limitation.