Optimize RVV affine transforms
This PR optimize the RVV implementations of the affine transforms.
For the dense transform it adds special cases, when the entire inner loop fits into a vector register, which gave a small speedup.
The sparse transform was optimized to use ChunkSize=1, because RVV currently doesn't have a dpbusd like instruction and it makes sure we don't operate on zero inputs.
SpacemiT X100 benchmarks of the changes:
```py
# pyshbench orig XXX 30
dense: +0.0137x
sparse +0.2898x
both: +0.3204x
```
```py
# stockfish bench
X100: orig: 170061 nodes/second both: 206581 nodes/second
A100: orig: 112999 nodes/second both: 128794 nodes/second
```
closes https://github.com/official-stockfish/Stockfish/pull/6967
No functional change
Co-authored-by: wCgiLKcH <111658515+43eayl5@users.noreply.github.com>