Add a matmul test from int8, bf16

[karupayun]

In this PR we are adding a matmul test from int8, bf16. I had a few issues in the test so
I refactored the file a bit.

• First I included two new params:

  • Dot_out_dtype: So users of the test class can specify the type used internally in the dot,
    and not the one set by default given the two types. There are several restrictions for
    these types anyway.
  • C_dtype: The return type of the matmul.
    I included a few tests in the case of making a dot with two float16.

• I had to modify test_matmul to use small integers when testing with two float16 since torch
  used float32 internally in this case and we were having precision issues when comparing
  the results with triton in the case that dot_out_dtype was float16.

• I also needed to include torch.int8 in the possible datatypes.

Finally I tried to simplify a bit the logic of the matmul/test_matmul because after adding
these two parameters it was a bit hard to follow why we needed every part of the code, so
I included a type_preference_list for the allowed dot_out_dtype given the types of the operands
a and b.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[gflegar]

Thanks! Definitely much simpler and more versatile than it was before. But I think it can be even simpler and clearer - left some comments to do that.

[gflegar]

Generally LGTM.

A few unresolved comments regarding scaled integers, and the weird view thing, we can also discuss them offline if you want.

[karupayun]

This PR was divided between triton-lang#2768, triton-lang#2769 and triton-lang#2760. All of them are already merged.