ggml-cpu: clean up s390x SIMD

[taronaeo]

This Pull Request cleans up the s390x SIMD Vector Intrinsics syntax to match the new code for easier readability and understanding. No new feature introduction, but there should be slight performance improvements switching the horizontal summation to the optimised vec_hsum.

Verified that models F32, F16, Q8_0, Q5_1, Q5_0, Q4_1, Q4_0, Q6_K, Q4_K, Q3_K, IQ4_XS, IQ4_NL still work as intended.

Performance Benchmark

model	size	params	backend	threads	test	t/s master	t/s PR	speedup
granite 3B all F32	9.44 GiB	2.53 B	BLAS	8	pp512	82.72	82.44	1.00
granite 3B all F32	9.44 GiB	2.53 B	BLAS	8	tg128	4.43	4.39	0.99
granite 3B F16	4.72 GiB	2.53 B	BLAS	8	pp512	74.93	74.89	1.00
granite 3B F16	4.72 GiB	2.53 B	BLAS	8	tg128	3.12	3.05	0.98
granite 3B Q8_0	2.51 GiB	2.53 B	BLAS	8	pp512	81.99	81.81	1.00
granite 3B Q8_0	2.51 GiB	2.53 B	BLAS	8	tg128	13.81	13.75	1.00
granite 3B Q5_1	1.78 GiB	2.53 B	BLAS	8	pp512	81.46	81.53	1.00
granite 3B Q5_1	1.78 GiB	2.53 B	BLAS	8	tg128	18.33	18.18	0.99
granite 3B Q5_0	1.64 GiB	2.53 B	BLAS	8	pp512	82.07	82.21	1.00
granite 3B Q5_0	1.64 GiB	2.53 B	BLAS	8	tg128	20.24	20.25	1.00
granite 3B Q4_1	1.49 GiB	2.53 B	BLAS	8	pp512	82.27	82.41	1.00
granite 3B Q4_1	1.49 GiB	2.53 B	BLAS	8	tg128	21.8	22	1.01
granite 3B Q4_0	1.35 GiB	2.53 B	BLAS	8	pp512	82.3	82.22	1.00
granite 3B Q4_0	1.35 GiB	2.53 B	BLAS	8	tg128	22.57	22.32	0.99
granite 3B Q6_K	1.94 GiB	2.53 B	BLAS	8	pp512	82.02	81.9	1.00
granite 3B Q6_K	1.94 GiB	2.53 B	BLAS	8	tg128	17.49	17.51	1.00
granite 3B Q4_K - Medium	1.44 GiB	2.53 B	BLAS	8	pp512	82.31	82.34	1.00
granite 3B Q4_K - Medium	1.44 GiB	2.53 B	BLAS	8	tg128	23.55	21.34	0.91
granite 3B Q3_K - Medium	1.16 GiB	2.53 B	BLAS	8	pp512	82.44	82.57	1.00
granite 3B Q3_K - Medium	1.16 GiB	2.53 B	BLAS	8	tg128	21.19	19.92	0.94
granite 3B IQ4_XS - 4.25 bpw	1.30 GiB	2.53 B	BLAS	8	pp512	82.01	82.18	1.00
granite 3B IQ4_XS - 4.25 bpw	1.30 GiB	2.53 B	BLAS	8	tg128	24.11	22.14	0.92
granite 3B IQ4_NL - 4.5 bpw	1.37 GiB	2.53 B	BLAS	8	pp512	82.08	82.28	1.00
granite 3B IQ4_NL - 4.5 bpw	1.37 GiB	2.53 B	BLAS	8	tg128	20.13	19.77	0.98

build: 4d49427 (6405)

Note

Tests were conducted on an IBM z17 Mainframe with 40 IFLs (cores) and 128 GB Memory on a shared R&D LPAR.