Prefixed batching; Improved batcher/analyzer error handling

[austinbyers]

to: @chunyong-lin size: medium

Background

The introduction of yextend has caused the BinaryAlert batch analysis to be more error-prone. This PR improves error logging and handling.

Changes

• The analyzer logs the entire Dynamo record (instead of just the composite key) before saving. This will make it easier in the future to troubleshoot PutItem errors.
• The analyzer catches any exception when running yextend instead of a specific set. We keep finding new errors which can crop up. In addition, we log the yextend output if there was an error to make it easier to debug errors.
• The batcher can now be restricted to a specific prefix (instead of scanning the whole bucket)
• Adds missing test coverage around yextend results with and without exceptions
• Updates requirements.txt with their latest versions.

Testing

• CI
• Live test with lots of files in S3 followed by a batch operation

[coveralls]

Coverage increased (+0.08%) to 93.137% when pulling ebad12f100e7789f510ccedfca4c323be99a8cb1 on austin-error-handling into db857f354b9f2c8b6a155c3c75180a6f07e5d1dd on master.

[austinbyers]

1. Correct, we can only analyze the whole bucket or one specific prefix. Those are the only options supported by the list_bucket API. If a need arises in the future, we could support multiple prefixes, but that would be a fairly complex change.
2. The prefix is just the beginning of an S3 key. So for example, you could set OBJECT_PREFIX=carbonblack/ to only batch analyze the downloads from Carbon Black.