Scoring code needs to account for held-out workload performance

[fsschneider]

The benchmark rules describe that submissions also need to perform well on held-out workloads. This is currently not automatically accounted for in the scoring code.
Specifically, (quoting from the benchmark rules):
For a submission to receive a finite training time on a fixed workload, it needs to:

• Reach the validation target on the fixed workload within the maximum runtime.
• Reach the validation target fixed workload within 4x of the fastest submission.
• Reach the validation target on the held-out workload (corresponding to the fixed workload) within the maximum runtime.
• Reach the validation target on the held-out workload (corresponding to the fixed workload) within 4x of the fastest submission. To determine the fastest submission on a held-out workload, we only consider submissions that reached the target on the corresponding fixed workload. This protects us against extremely fast submissions that only work on a specific held-out workload and are useless as general algorithms.

[priyakasimbeg]

Remaining work on this is to add tests that check if all 4 criteria are enforced in strict mode.

[priyakasimbeg]

Moved tracking of tests to #624