Speed up CI without reducing coverage

[acgetchell]

Summary

Apply the low-risk CI speedups that preserve coverage and correctness while reducing repeated Cargo work.

This issue tracks the equivalent of the Delaunay follow-up items 3, 4, and 6:

• Build examples once, then run the compiled example binaries instead of invoking cargo run for each example.
• Keep feature-gated bench/example lint coverage, but avoid unnecessarily monolithic duplicate Clippy target graphs where possible.
• Use cargo nextest for Rust test execution wherever it can preserve behavior and coverage.

Constraints

• Do not reduce test, example, or benchmark compile coverage.
• Preserve feature-gated coverage for any bench/example paths.
• Keep local comprehensive validation commands comprehensive.
• Keep cargo test for doctests and for any exact/debug flow where nextest cannot preserve behavior cleanly.
• Measure before/after timing where practical.

Acceptance criteria

• Examples, if present, are compiled once and then executed directly.
• Clippy coverage still includes all relevant target classes and feature-gated code.
• Rust test recipes use cargo nextest wherever possible, with remaining cargo test uses documented as doctest-only or behavior-preserving exceptions.
• Integration tests, if present, use nextest or have a documented reason not to.
• The full local/CI validation recipe passes.