Reposition README around lint-time checking and coverage gate, refresh benchmarks

[w-martin]

Summary

• Repositions the README's hero example and "Why typedframes?" section around how the tool is mainly used today: a zero-schema Rust linter plus an opt-in DataFrame schema coverage threshold for CI (analogous to a test-coverage gate or a bandit-style security lint gate), rather than leading with BaseSchema schema definitions. Schema classes are now framed as a progressive enhancement (cross-file awareness, IDE autocomplete) and moved into Quick Start.
• Promotes "DataFrame Schema Coverage Thresholds" from a subsection of "Static Analysis" to its own top-level section, with a Table of Contents entry.
• Re-runs benchmarks/benchmark_checkers.py --update-readme for current tool versions and timings (ruff 0.16.2, ty 0.0.69, pyrefly 1.2.0, mypy 2.3.0; pyright already current).
• Fixes stale/inaccurate claims found while verifying against source and the live benchmark run:
  • --json example replaced with --output-format=json (the documented flag; --json is a hidden legacy alias in the CLI).
  • The "Standalone checker ✅ Rust (~1ms)" comparison-table cell and "ty ... 10-60x faster than mypy" claim no longer matched measured numbers — replaced with wording that matches the actual benchmark data.
  • Stray tool-version mentions elsewhere in prose synced to the refreshed benchmark table.

Test plan

• uv run pytest -q — 299 passed
• Verified every internal #anchor link in the README resolves to a real heading
• Verified every --flag mentioned in the README exists in typedframes check --help
• Verified comparison-library version pins (Pandera, Great Expectations, etc.) already match their uv.lock files — no change needed there