vulkan: optimize iq3_xxs mat-vec shader

[WakeUpMorty]

Rewritten mul_mat_vec_iq3_xxs mat-vec shader. Kernel goes ~65% → 74% of peak
bandwidth, total gain ~+0.5%. Related: #28426 (iq4_xs, filed together; no file
overlap)

Overview

Optimized the mul_mat_vec_iq3_xxs Vulkan compute shader for the dmmv path by mirroring the approach used in mul_mat_vec_iq3_s.
Instead of using two invocations per 32-value subblock and loading the grid via unaligned 1-byte accesses,
the new version assigns one invocation per subblock (totaling 8 per 256-value block). It now loads the grid using aligned u16 packed16 memory accesses.
The underlying math and accumulation order remain completely unchanged.

Benchmark (Qwen3.8-27B Q3_K_XL, llama-bench -r 3, tg leg d0, fresh context)

Reference = official prebuilt 10791.
Patched = 427291b + the iq4_xs PR + this change (this change sits on top of the iq4_xs PR).

build	tg256	tg512
reference 10791 (official prebuilt)	34.10	34.00
patched (iq4_xs PR + this change)	36.95	36.97
combined gain	+8.4%	+8.7%

Standalone effect (perf logger, 32 tokens):

build	iq3_xxs eff	iq3_xxs ms/tok
official 10791	451 GB/s	2.245
final build	471 GB/s	2.146

= ~0.5% of total token time.

Per-shape iq3_xxs efficiency (GB/s, old → new):

shape	old	new	(iq3_s at same shape)
m=17408 k=5120	454	531	665
m=12288 k=5120	403	482	614
m=10240 k=5120	391	452	607
m=5120 k=17408	340	403	601
m=5120 k=6144	275	372	551
m=6144 k=5120	263	336	504

Correctness: bit-identical output at temperature 0 (64 tokens, same prompt,
all three bench models) on the final combination build (both PRs applied).

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES - AI was used to help with shader development, benchmarking, and this PR description. All changes were reviewed and verified by the author.

[ggml-gh-bot]

Hi @WakeUpMorty, thanks for your contribution!

Per our contribution guidelines, the automated PR checker found the following issue(s) that need your attention:

• Multiple open PRs from a new contributor: We limit new contributors (those without a previously merged PR) to 1 open PR at a time. You currently have 2 open PRs.

• AI-generated content: While code is allowed to be generated by AI, please write the PR description and commit messages on your own without the help of AI.

---

Please note that maintainers reserve the right to make final decisions on PRs. If you believe there is a mistake, please comment below.

[WakeUpMorty]

Rewrote the description. Could the check be re-run?