ggml-cpu(s390x): add Q1_0 vector intrinsic support

[taronaeo]

Overview

This PR introduces vector intrinsic support for Q1_0 data type. We are looking at performance improvements of approximately 139.53% for Prompt Processing and 131.59% for Token Generation.

Additional information

Users of the Bonsai 1-bit models are recommended to compile with -DGGML_BLAS=ON to achieve higher performance.

Performance Benchmark

$ build/bin/llama-bench -hf taronaeo/Bonsai-8B-BE-GGUF:Q1_0 -lm mlock

Before

model	size	params	backend	threads	lm	test	t/s
qwen3 8B Q1_0	1.07 GiB	8.19 B	CPU	2	mlock	pp512	0.52 ± 0.00
qwen3 8B Q1_0	1.07 GiB	8.19 B	CPU	2	mlock	tg128	0.46 ± 0.00

build: ca86fb2 (10859)

After

model	size	params	backend	threads	lm	test	t/s
qwen3 8B Q1_0	1.07 GiB	8.19 B	CPU	2	mlock	pp512	2.92 ± 0.01
qwen3 8B Q1_0	1.07 GiB	8.19 B	CPU	2	mlock	tg128	2.23 ± 0.01

build: 6577365 (10861)

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: Yes, Opus 5 because my initial idea was poor for performance

[taronaeo]

cc: @AlekseiNikiforovIBM @Andreas-Krebbel for review if any

[AlekseiNikiforovIBM]

LGTM