[Benchmark Script] Refactor benchmark script for `bench_datatype_gemm`

[yewentao256]

I know this is a bnechmark script but I think this could still be refactored. Perhaps a few functions/objects and a dictionary?

Originally posted by @ProExpertProg in #19233 (comment)

I think this is a great idea, example config designed:

PROVIDER_CFGS = {
    "int8-tensor-w-token-a": dict(w="tensor", a="token", no_a_quant=False),
    "int8-tensor-w-tensor-a": dict(w="tensor", a="tensor", no_a_quant=False),
    "int8-channel-w-token-a": dict(w="channel", a="token", no_a_quant=False),
    "int8-channel-w-tensor-a": dict(w="channel", a="tensor", no_a_quant=False),
    "int8-tensor-w-token-a-noquant": dict(w="tensor", a="token", no_a_quant=True),
    "int8-tensor-w-tensor-a-noquant": dict(w="tensor", a="tensor", no_a_quant=True),
    "int8-channel-w-token-a-noquant": dict(w="channel", a="token", no_a_quant=True),
    "int8-channel-w-tensor-a-noquant": dict(w="channel", a="tensor", no_a_quant=True),
}

After #19233 merged, I can have another PR optimizing this, and update benchmarks/kernels/bench_fp8_gemm.py as well.

[ProExpertProg]

Nice! I think PROVIDER_CFGS should also contain a lambda/function ref that executed that benchmark if possible