[RVV]: Add RVV xx-fill kernel

[leduclean]

Summary

• Add xnn_xx_fill_ukernel__rvv_u4v, an RVV implementation of the xx-fill microkernel (fills a buffer with a repeating 32-bit pattern) using LMUL=4, registered in the xx-fill config/build files.
• Add bench/xx-fill.cc (modeled on bench/xx-pad.cc) to benchmark it against the scalar_u16 fallback.

LMUL tuning

Tuned empirically on k230 across m8/m4/m2:

• m4 matches m8's peak throughput on large shapes while fixing m8's only weak spot (small channel counts).
• m2 trades away large-shape throughput (-21% to -37%) for a bit more on the smallest shape only, a bad trade, not used.

Perf

Measured on k230 (single core, 27 MHz), bytes/s, rvv (m4) vs scalar_u16:

rows	channels	rvv	scalar	speedup
1	16	1.00 G/s	806.0 M/s	1.24x
1	64	4.00 G/s	1.22 G/s	3.28x
1	256	8.16 G/s	1.40 G/s	5.84x
1	1024	11.03 G/s	1.45 G/s	7.61x
1	8192	12.20 G/s	1.47 G/s	8.32x
100	256	11.94 G/s	1.45 G/s	8.25x
1024	16	2.47 G/s	1.25 G/s	1.98x

[leduclean]

Thanks for the review!
All comments addressed, PTAL.

[leduclean]

Hello @qukhan, same as #10917 this was approved on Aug 17 but hasn't landed yet. I've rebased it on master. Anything else needed on my side?