IoT optimised ML-KEM

[franziskuskiefer]

Implement an IoT optimised version of ML-KEM by applying optimisations specific to the reference board.

To measure performance, we should have numbers for

• Benchmarks (time)
• Stack usage
• Binary size

Our reference board has a Cortex-M4F, so all optimizations are a priori specific to that platform.

[franziskuskiefer]

@jschneider-bensch please add sub-issues for the individual items you're tackling so that we can track how far we are and what's working, and what not.