[Perf] Use a separate parallel job for each set of test arguments

Expanding on #3549, we could go further and use a separate parallel job for each set of test arguments. So storage-blob would use a total of 11 parallel jobs (4 for download, 4 for upload, and 3 for list-blobs).

This may use too many agents as we scale up the number of tests and arguments. And there is a point where too much parallelization wastes too much time cloning and deploying resources to only run a single test. But it would give developers the quickest results when running ad-hoc perf tests.

This might be a dial we can turn, where for scheduled runs we reduce parallelism to save resources, and for ad-hoc runs we enable max parallelism.