add API coverage test

[FindHao]

This PR adds the userbenchmark for PyTorch API coverage test

Command to run for a specific model:

python3 run_benchmark.py api-coverage --model resnet18  --device cuda --test train,eval

If you want to test all models, the --model should not be specified.

The testing results are stored in .userbenchmark/api-coverage/logs/. Each run will generate two log files, logs-${timestamp}.json and logs-${timestamp}.json-api_coverage.csv. The latter one is like the following format.

API coverage rate: 25/1332 = 1.88%
=====Used APIs=====
module_name,func_name
torch,flatten
torch._C._nn,linear
...
=====Missed APIs=====
module_name,func_name
,_TensorBase.acos
...

All Used and missed APIs will be listed in the csv file.

[xuzhao9]

@FindHao Can you update the PR body to the command of running the new userbenchmark and its example outputs?

[facebook-github-bot]

@xuzhao9 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[FindHao]

@FindHao Can you update the PR body to the command of running the new userbenchmark and its example outputs?

Done.

[FindHao]

ping @xuzhao9

[facebook-github-bot]

@FindHao has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[facebook-github-bot]

@FindHao merged this pull request in 09de70c.