- Used when non-constant indices are given to Shuffle, and an intrinsic implementation of ShuffleUnsafe is available
- Optimises byte and sbyte cases only

- Re-implement ShuffleUnsafe in JIT
- Also re-implement the Shuffle optimisations in JIT
- Add basic JIT support for mono (ShuffleUnsafe just gets the same implementation as Shuffle)
- Implement support for variable index Shuffle & ShuffleUnsafe (for bytes)
- Implement support for cross-lane shuffling in JIT (for bytes)
- Optimise Vector128 shuffle for bytes in JIT to use Avx2.Shuffle

- Fix incorrect usage of gtNewConditionalNode
- Fix type of valueMask
- Add assert
- Use gtNewSimdCreateBroadcastNode where possible

- Logic change for short shuffle was incorrect: reverted
- Accidentally ignored permutation node when shuffling also
- Read from op2 instead of vecCns.u8, but assumed values were in the 0->31, 255 ranges

- Use Avx2.Shuffle for contant-index short shuffle operations too
- Implement cross-lane shuffles for constant-index short shuffles

- Implement `ShuffleUnsafe` for `sbyte`
- Remove remaining detailed comment on V128.ShuffleUnsafe (which specified which intrinsics were used): it is not good to only have it for this one, which could also theoretically change behaviour in the future e.g., with the introduction of RISC-V

- Implement non-constant indices handling for Shuffle & ShuffleUnsafe for all type sizes on arm64

- Implement non-constant indices handling for Shuffle & ShuffleUnsafe for all type sizes on xarch