Use slow tests stats in common_utils (#55190)
Summary:
This is a step in adding automatic slowTest detection to our testing infrastructure. This uses stats (updated daily) in https://github.com/pytorch/test-infra/blob/master/stats/.pytorch-slow-tests to determine whether more tests need to be marked as slow as they are run.
More details in previous PR draft/proposal [here](https://github.com/pytorch/pytorch/pull/54456#issue-598388491), though I no longer think we need the third step as using a raw git file does not require much processing.
Upon looking at [logs](https://circleci.com/api/v1.1/project/github/pytorch/pytorch/12060292/output/107/0?file=true&allocation-id=606660dbd8e5857bcc2b2e0f-0-build%2F60DCA8CD) for the coverage tests as of the first commit [when I had not skipped the tests so we could see their actual times], here are some slow tests that weren't marked as slow before:
```
test_fn_gradgrad_unfold_cpu_complex128 (__main__.TestGradientsCPU) (172.554s)
test_matmul_4d_4d_complex_cpu (__main__.TestAutogradDeviceTypeCPU) (180.057s)
test_conv1d_basic (__main__.TestXNNPACKConv1dTransformPass) (94.737s)
```
And here is a test that wasn't actually slow but was still marked as slow based on stats:
```
test_trunc_normal (__main__.TestNNInit) ... ok (1.208s)
```
The new logs show the above tests as skipped (as they should be):
[Coverage Test 1](https://app.circleci.com/pipelines/github/pytorch/pytorch/296224/workflows/ba6c2917-51f8-4fb8-be57-90151c2e5c25/jobs/12126156) and [Coverage Test 2](https://app.circleci.com/pipelines/github/pytorch/pytorch/296224/workflows/ba6c2917-51f8-4fb8-be57-90151c2e5c25/jobs/12126155)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55190
Reviewed By: samestep
Differential Revision: D27566663
Pulled By: janeyx99
fbshipit-source-id: c13f8c676bb8eb15d9d697d224dbaef7df98aef3
