Test all implementation variants in CI

[mkannwischer]

We have multiple implementations variants for Keccak now.
For NTTs, we currently have C and asm as well now, and we'll see mulitple asm variants in the future.

Currently, for each platform exactly one (ideally the fastest) is tested.
CI should be extended to test all (functional) variants on all platforms -- not that not all platforms can run all implementations - for example, SHA-3 instructions are not available on platforms < v8.4.
Benchmarks should continue to be limited to only the fastest implementation.