Add Support for different GGML_TYPE inputs

[MaoxinYee]

ggml_tensor * ggml_mul_mat(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * b)

If a->type== GGML_TYPE_Q8_0 and b->type == GGML_TYPE_F32:

ggml_hsa_compile_kernel: failed to compile kernel mul_mat-128256x512f32-4096x128256q8_0-4096x512f32 for tensor "" (MUL_MAT): ValueError: Input dtype (<class 'numpy.int8'>) and output dtype (<class 'numpy.float32'>) must either both be integral or both be float