Vulkan: add GGML_OP_FUSED_MUL_UNARY

[ikawrakow]

The tiniest of performance increases, barely measurable.

But now we no longer need to special-case Vulkan when building the graph.

Of note: I went to measure mainline llama.cpp Vulkan performance with my setup (RTX-4080 in a Ryzen-5975WX box, Nvidia driver 575, so coopmat2 enabled). Interestignly enough, ik_llama.cpp is 10-15% faster than mainline for a context of 16k tokens, even though there aren't any noticeable Vulkan optimizations in ik_llama.cpp yet. Is mainline paying the price for the grand KV cache unification?

Here sweep-bench results for LlaMA-3.1-8B-Instruct

Mainline llama.cpp (build: 5821)

PP	TG	N_KV	T_PP s	S_PP t/s	T_TG s	S_TG t/s
1024	256	0	0.244	4200.63	2.742	93.37
1024	256	1024	0.256	4002.55	2.730	93.78
1024	256	2048	0.282	3634.36	2.827	90.57
1024	256	3072	0.297	3452.54	2.897	88.36
1024	256	4096	0.319	3212.57	2.956	86.62
1024	256	5120	0.335	3056.04	3.045	84.08
1024	256	6144	0.349	2937.19	3.126	81.90
1024	256	7168	0.365	2807.69	3.184	80.40
1024	256	8192	0.378	2710.13	3.284	77.97
1024	256	9216	0.396	2589.00	3.364	76.10
1024	256	10240	0.407	2514.06	3.453	74.13
1024	256	11264	0.424	2415.06	3.518	72.77
1024	256	12288	0.441	2322.30	3.621	70.70
1024	256	13312	0.455	2249.22	3.704	69.12
1024	256	14336	0.468	2190.30	3.786	67.62
1024	256	15360	0.485	2111.54	3.852	66.45

ik_llama.cpp with this PR

PP	TG	N_KV	T_PP s	S_PP t/s	T_TG s	S_TG t/s
1024	256	0	0.242	4234.03	2.673	95.77
1024	256	1024	0.253	4052.19	2.646	96.74
1024	256	2048	0.265	3866.73	2.718	94.20
1024	256	3072	0.283	3618.48	2.775	92.25
1024	256	4096	0.286	3584.28	2.830	90.45
1024	256	5120	0.298	3441.80	2.905	88.13
1024	256	6144	0.321	3194.99	2.983	85.81
1024	256	7168	0.335	3059.90	3.034	84.37
1024	256	8192	0.340	3007.63	3.089	82.87
1024	256	9216	0.353	2897.76	3.128	81.83
1024	256	10240	0.364	2814.86	3.192	80.21
1024	256	11264	0.372	2753.81	3.241	78.99
1024	256	12288	0.380	2692.31	3.291	77.78
1024	256	13312	0.397	2580.87	3.370	75.97
1024	256	14336	0.412	2486.45	3.444	74.33
1024	256	15360	0.423	2420.49	3.498	73.19