test_import_stats is flaky

[malfet]

test_import_stats.py accidentally fails during call to register_rds_schema:
Traceback (most recent call last):
File "test_import_stats.py", line 63, in
"time_ms": "int",
File "/var/lib/jenkins/workspace/tools/stats/scribe.py", line 94, in register_rds_schema
invoke_rds(event)
File "/var/lib/jenkins/workspace/tools/stats/scribe.py", line 72, in invoke_rds
return invoke_lambda("rds-proxy", events)
File "/var/lib/jenkins/workspace/tools/stats/scribe.py", line 33, in invoke_lambda
raise Exception(payload)
Exception: {"errorMessage":"2021-10-06T01:14:28.434Z ae80dd87-8803-4ab6-843b-08f1011c22a3 Task timed out after 3.00 seconds"}

For example for https://hud.pytorch.org/commit/83bac89d6477d084ac4020de56c13ecc42e0f7ac it failed for 3 jobs:
https://github.com/pytorch/pytorch/runs/3809711391#step:16:23571
https://github.com/pytorch/pytorch/runs/3809744116#step:16:3818
https://github.com/pytorch/pytorch/runs/3809744150#step:16:3822

Another fun fact: this kind of error does not register in generated test_results.xml

cc @ezyang @gchanan @zou3519 @bdhirsh @jbschlosser @seemethere @malfet @pytorch/pytorch-dev-infra

[malfet]

One more failure in ONNX stack: https://github.com/pytorch/pytorch/runs/3809277548

[driazati]

it's not in the test report since the failing code is not in an actual test (it happens before unittest.main()), I'll follow up with a PR to change how the API works there (so you can idempotently register during tests) and move some retry/error ignoring lower since this should never really show up as an actual failure