[arthw]

Implement the feature request: #21641

support MUL_MAT and OUT_PROD with Q1_0.
all related UT cases are passed.

Test LLM: Bonsai-1.7B-Q1_0.gguf

on B60

Test	Flash Attn	pub build (t/s)	PR build (t/s)	Speedup
pp512	fa = 0	155.23 ± 0.40	3041.99 ± 4.22	19.6x
pp512	fa = 1	153.45 ± 0.06	2685.07 ± 1.75	17.5x
tg128	fa = 0	60.16 ± 0.12	148.47 ± 0.13	2.47x
tg128	fa = 1	65.74 ± 0.07	159.67 ± 0.54	2.43x

on Xe GPU in Ultra 7 265K

Test	Flash Attn	pub build (t/s)	PR build (t/s)	Speedup (PR vs pub)
pp512	fa = 0	131.42 ± 0.34	472.65 ± 0.06	3.60x
pp512	fa = 1	114.29 ± 0.14	291.50 ± 0.05	2.55x
tg128	fa = 0	13.12 ± 0.00	16.67 ± 0.00	1.27x
tg128	fa = 1	16.25 ± 0.01	21.83 ± 0.00	1.34x

On UHD 770 in i7-13700K

Test	fa	Base t/s	Primary t/s	Increase Rate (Primary vs Base)
pp512	0	196.56	212.32	8.02%
pp512	1	122.01	129.45	6.10%
tg128	0	24.42	18.39	-24.69%
tg128	1	27.26	19.23	-29.46%

In base, 400% CPU usage, PR use 100% CPU.
the result of base comes from CPU, instead of GPU.
So UHD 770 replace 3 CPU cores in above result.