Improve performance of stable sort

This reworks the internals of slice::sort. Mainly:
- Introduce branchless swap_next_if and optimized sortX functions
- Speedup core batch extension with sort16
- Many small tweaks to reduce the amount of branches/jumps

This commit is incomplete and MUST NOT be merged as is. It is missing Copy
detection and would break uniqueness preservation of values that are being
sorted.

Author: Voultapher