Integration tests are flaky -- replace dataset sampling with top 50 datasets

[betolink]

Right now the integration tests use randomly sampled datasets from different DAACs, this results in intermittent failures due EULAs/unavailability etc. A probably better approach would be to test only the most popular datasets (+IS2 data). A list of the most popular datasets from NASA is attached.
collection_short_names.txt

[MattF-NSIDC]

Maybe we could keep the random-sample test as an optional test suite? It would be useful to keep track of corner cases, but we wouldn't really want our main regression test suite to fail because of a new corner case.

[asteiker]

@mfisher87 we're leaving this open for now knowing that it's likely resolved with #749. Are you still needing reviewers to move the PR forward? Lmk how I can help facilitate!

[mfisher87]

We do need more work on that PR. It's been tough moving it forward just during hack days! I think the remaining tasks are achievable without needing decision input, just requires some time.