Improve precision recall metric issue 2571

[sadra-barikbin]

Fixes #2571

Check list:

• New tests are added (if a new feature is added)
• New doc strings: description and/or example code are in RST format
• Documentation is updated (if required)

[vfdev-5]

@sadra-barikbin thanks a lot for the PR !
I left few comments to clear out the API. I haven't checked yet the details but the feature looks cool !
Please check also docs build failure.

[sadra-barikbin]

Is there any reason that this test has not been parametrized?

[vfdev-5]

This test and other similar ones are used in test_distrib_hvd, test_distrib_gloo_cpu_or_gpu, test_distrib_nccl_gpu tests. IMO, it is not obvious how to properly parametrize them. However, there was a task to rewrite distributed tests such that we do not need to write all: test_distrib_hvd, test_distrib_gloo_cpu_or_gpu, test_distrib_nccl_gpu but only _test_distrib_integration_multiclass and they could be executed with hvd, gloo, nccl confs...

[sadra-barikbin]

Scikit-Learn has another data type called multilabel-multioutput which is like multilabel but its labels are not binary. It has yet to be implemented. We can do it if you will.

[vfdev-5]

Thanks a lot for this great PR @sadra-barikbin !
I'm checking it on GPU and if CI is green let's merge it.
In a follow-up PR, let's clean-up docstring about

- torch.Tensor(...)
+ torch.tensor(...)