[FIX] Fix cublas batch matmul

[sxjscience]

Thanks for contributing to TVM! Please refer to guideline https://tvm.apache.org/docs/contribute/ for useful information and tips. After the pull request is submitted, please request code reviews from Reviewers by @ them in the pull request thread.

@comaniac The issue is the same as the cblas_batch_matmul issue in #6699.

[comaniac]

@merrymercy the CI failed at the Ansor tutorial. Would you mind take a look?

[icemelon]

Why do we need this attribute in batch_matmul?

[sxjscience]

The reason is that the code here always assume that there is an out_shape argument. https://github.com/apache/incubator-tvm/blob/461e75bd5ffaf45a0f270998514d444463d11261/python/tvm/relay/op/strategy/generic.py#L685-L686

[icemelon]

I checked the implementation. I think we can remove the oshape arg in the topi implementation and we can enable broadcasting at batch axis by default. cc @jwfromm

https://github.com/apache/incubator-tvm/blob/main/python/tvm/topi/nn/batch_matmul.py#L23

[sxjscience]

Per offline discussion with @icemelon9 , I will have a separate PR to remove the oshape flag directly.

[icemelon]

LGTM

[icemelon]

Thanks @sxjscience @comaniac