now we've had simd optimization in Q8_0 and Q8_1, we can add more optimizations over the basic quantization formats (not include the K quants).

we can take a new approach on making SIMD optimizations:

1. add a cross-platform SIMD optimization via stdsimd, and name it vec_dot_xxx_xxx_fallback, which offers a not too bad performance over different platforms.
2. add a neon optimized vec_dot implementation, which optimized on apple silicon platform
3. add a x86_64 platform with avx2

we do not need add platform specialized code in the quantize and dequantize part, using stdsimd or even plain rust loops is ok, it does not actually affect performance while comparing with a platform specialized impl.

• simd in Q8_0, Q8_1
• NEON SIMD in Q4_0 #159
• NEON/AVX2 SIMD in Q4_1 #160
• AVX2 SIMD in Q4_0 #174
• AVX2 SIMD in Q5_0 #161
• AVX2 SIMD in Q5_1 #162
• Support RISC-V SIMD in Q8_0 #169
• Neon SIMD in Q5_0 #170
• NEON SIMD in Q5_1 #171
• Support RISC-V SIMD in Q4_0 #172
• Support RISC-V SIMD in Q4_1 #173