We handle test reports in different ways:

1. We send them to S3 in their raw form here
2. We send them to S3 (a different bucket) in a parsed form in print_test_stats.py
3. We use the raw form for PR HUD, like https://hud.pytorch.org/commit/f767cf668395baf29ca7c9f1fa80f0abed8c53c7
4. We use parsed form for test categorization like with slow tests and sharding and target det tests.

When steps 1 or 2 fail, they can affect 3 and 4, which affect developer experience. We would like to detect this failures early on rather than later down the line. We also want to be able to log these processes for ease of debugging.

@suo @pytorch/pytorch-dev-infra