docs: simplify performance guidance and centralize benchmark snapshots

[dloss]

Motivation

• The previous performance docs were long, duplicated raw benchmark tables, and aged quickly, making guidance hard to maintain.
• The goal is to provide short, evergreen guidance for tool choice and tuning while preserving reproducible raw snapshots separately.

Description

• Replace docs/concepts/performance-comparisons.md with a concise decision guide that includes a TL;DR tool-selection matrix, honest performance positioning, and a fast-path tuning checklist.
• Rework docs/concepts/benchmark-results.md into a raw-snapshot hub with a canonical snapshot template and instructions to treat benchmarks/comparison_results/ as the source of truth.
• Remove large embedded benchmark tables and examples from the guidance page and instead point readers to generated artifacts for per-machine numbers.
• Update cross-links and small formatting tweaks so guidance and snapshot pages are consistent with the two-layer model (guidance vs raw artifacts).

Testing

• Ran code formatting via cargo fmt --all, which completed successfully.
• Ran linting via cargo clippy --all-targets --all-features -- -D warnings, which completed successfully with no warnings.
• Ran the test suite via cargo test, which completed successfully (1047 tests passed).
• Attempted just fmt but the just tool is not installed in this environment, so that external check could not be executed here.

---

Codex Task