Try compilation settings that might speed up execution

[teor2345]

This work is blocked by having a flamegraph (#39) and a working full implementation (so we know which functions are hot).

Benchmark these settings against each other:

• [ ] levels 2, 3 (default), s, z optimisation
• [ ] #[inline(always)] for small hot methods (this might need #39)
• [ ] https://nnethercote.github.io/perf-book/build-configuration.html#alternative-allocators
• [ ] https://nnethercote.github.io/perf-book/build-configuration.html#profile-guided-optimization
• [ ] Double-check the settings in #42 actually improve performance

This is potentially a blocker for splitting out generic polynomial operations.