ggml-cpu: faster AVX2 variant for IQ1_M

[remyoudompheng]

This is an additional optimization for AVX2 after #12154

Benchmarks on AMD Zen 4:

Before
  MUL_MAT(type_a=iq1_m,type_b=f32,m=4096,n=1,k=14336,bs=[1,1],nr=[1,1],per=[0,1,2,3]):                  1932 runs -   532.51 us/run - 117.44 MFLOP/run - 220.54 GFLOPS
  MUL_MAT(type_a=iq1_m,type_b=f32,m=4096,n=2,k=14336,bs=[1,1],nr=[1,1],per=[0,1,2,3]):                   980 runs -  1057.22 us/run - 234.88 MFLOP/run - 222.17 GFLOPS

After
  MUL_MAT(type_a=iq1_m,type_b=f32,m=4096,n=1,k=14336,bs=[1,1],nr=[1,1],per=[0,1,2,3]):                  2553 runs -   394.78 us/run - 117.44 MFLOP/run - 297.48 GFLOPS
  MUL_MAT(type_a=iq1_m,type_b=f32,m=4096,n=2,k=14336,bs=[1,1],nr=[1,1],per=[0,1,2,3]):                  1225 runs -   816.50 us/run - 234.88 MFLOP/run - 287.67 GFLOPS

model	size	params	ngl	test	master t/s	PR t/s
qwen2 7B IQ1_M - 1.75 bpw	1.90 GiB	7.62 B	0	pp512	204.84 ± 0.31	204.07 ± 0.58
qwen2 7B IQ1_M - 1.75 bpw	1.90 GiB	7.62 B	0	tg128	14.48 ± 0.02	17.71 ± 0.18

[slaren]

13900k:

Model	Threads	Test	t/s master	t/s optim-x86-2	Speedup
llama 8B IQ1_M - 1.75 bpw	8	pp128	20.85	25.11	1.20
llama 8B IQ1_M - 1.75 bpw	8	tg32	16.34	19.54	1.20
llama 8B IQ1_M - 1.75 bpw	16	pp128	29.23	35.90	1.23
llama 8B IQ1_M - 1.75 bpw	16	tg32	20.84	24.17	1.16
llama 8B IQ1_M - 1.75 bpw	24	pp128	32.41	39.21	1.21
llama 8B IQ1_M - 1.75 bpw	24	tg32	22.41	25.95	1.16
llama 8B IQ1_M - 1.75 bpw	32	pp128	37.96	45.32	1.19
llama 8B IQ1_M - 1.75 bpw	32	tg32	22.78	26.94	1.18