Add NEON for highbd convolve_2d_sr, convolve_x_sr, convolve_y_sr

[jjustiss-apple]

Add NEON intrinsics for av2_highbd_convolve_2d_sr, convolve_x_sr, and
convolve_y_sr. The 2D kernel uses lane-indexed multiply-accumulate with a
register-tiled sliding-window vertical pass. The x_sr and y_sr kernels use
lane-indexed widening MAC to compute 8 output pixels simultaneously.

Micro-benchmark results (Apple Silicon M2, 10-bit, bench_kernels):

Kernel	8x8	16x16	32x32	64x64
convolve_x_sr	6.61x	8.32x	9.80x	9.47x
convolve_y_sr	11.72x	15.54x	16.18x	15.90x
convolve_2d_sr	10.30x	11.94x	12.15x	12.37x

CTC results (RA, cpu-used=1, 33 frames, A5, ARM Apple Silicon):

Metric	Delta
Encode time	-7.1%
Decode time	-11.3%
BD-rate Y	0.000%
BD-rate Cb	0.000%
BD-rate Cr	0.000%

Unit tests included (BD 8/10/12 coverage for all three functions).