sycl : Implemented reorder Q4_K mmvq

[sgeor255]

This PR enables reorder optimization for Q4_K layout similarly to #12858 . This branch is based off of @Alcpz 's and before that is merged the easiest way to review it is looking at the diff for 8cbe2c9 .

Some performance numbers on lunar lake below:

• Q4_K reorder with GGML_SYCL_DISABLE_OPT=0

model	size	params	backend	ngl	threads	sm	test	t/s
qwen2 1.5B Q4_K - Medium	1.04 GiB	1.78 B	SYCL	99	8	none	pp512	1586.19 ± 69.35
qwen2 1.5B Q4_K - Medium	1.04 GiB	1.78 B	SYCL	99	8	none	tg128	41.23 ± 0.43
llama 7B Q4_K - Medium	3.80 GiB	6.74 B	SYCL	99	8	none	pp512	550.65 ± 1.35
llama 7B Q4_K - Medium	3.80 GiB	6.74 B	SYCL	99	8	none	tg128	17.67 ± 1.05
gemma2 2B Q4_K - Medium	1.59 GiB	2.61 B	SYCL	99	8	none	pp512	616.41 ± 12.21
gemma2 2B Q4_K - Medium	1.59 GiB	2.61 B	SYCL	99	8	none	tg128	28.57 ± 0.32
llama 8B Q4_K - Medium	4.58 GiB	8.03 B	SYCL	99	8	none	pp512	508.14 ± 1.50
llama 8B Q4_K - Medium	4.58 GiB	8.03 B	SYCL	99	8	none	tg128	13.75 ± 0.12
phi3 3B Q4_K - Medium	2.23 GiB	3.82 B	SYCL	99	8	none	pp512	827.73 ± 26.59
phi3 3B Q4_K - Medium	2.23 GiB	3.82 B	SYCL	99	8	none	tg128	21.45 ± 0.17

build: 52b1622 (5099)

• Q4_K reorder with GGML_SYCL_DISABLE_OPT=1

model	size	params	backend	ngl	threads	sm	test	t/s
qwen2 1.5B Q4_K - Medium	1.04 GiB	1.78 B	SYCL	99	8	none	pp512	1576.79 ± 80.93
qwen2 1.5B Q4_K - Medium	1.04 GiB	1.78 B	SYCL	99	8	none	tg128	36.27 ± 0.43
llama 7B Q4_K - Medium	3.80 GiB	6.74 B	SYCL	99	8	none	pp512	551.82 ± 1.63
llama 7B Q4_K - Medium	3.80 GiB	6.74 B	SYCL	99	8	none	tg128	12.24 ± 1.19
gemma2 2B Q4_K - Medium	1.59 GiB	2.61 B	SYCL	99	8	none	pp512	586.64 ± 1.65
gemma2 2B Q4_K - Medium	1.59 GiB	2.61 B	SYCL	99	8	none	tg128	24.04 ± 0.41
llama 8B Q4_K - Medium	4.58 GiB	8.03 B	SYCL	99	8	none	pp512	509.51 ± 0.87
llama 8B Q4_K - Medium	4.58 GiB	8.03 B	SYCL	99	8	none	tg128	10.18 ± 0.04
phi3 3B Q4_K - Medium	2.23 GiB	3.82 B	SYCL	99	8	none	pp512	825.29 ± 26.93
phi3 3B Q4_K - Medium	2.23 GiB	3.82 B	SYCL	99	8	none	tg128	17.83 ± 0.05

build: 52b1622 (5099)

• TODO
  • Performance on BMG and ARC

[qnixsynapse]

Question: Is there a specific reason data_device is declared as a char* instead of a uint8_t*, especially considering it's later cast to uint8_t* as qs_ptr anyway?