Enhance log classifier to capture multiple errors in keep-going mode

[huydhn]

This is the next task for log classifier after pytorch/pytorch#138541 (comment). One of the big limit of the log classifier is that it could only find the last failures in the log. This simply doesn't work in keep-going mode in which testing will continue even if there are failures, and will only fail at the end.

Notes to bootcampers:

• Log classifier code is at https://github.com/pytorch/test-infra/tree/main/aws/lambda/log-classifier
• We need to update the schema of torchci_classification on workflow_job ClickHouse table to store the information
• A bonus point is to make this work with Bedrock prompt too https://github.com/pytorch/test-infra/blob/main/aws/lambda/log-classifier/src/bedrock/prompts.rs