This matches the old aitf-mirror behavior (and matching tests)

Relative to the dst passed to Downloader's update_cache/download_for_time

It was an overly clever hack. Manage summaries directly

Previously we relied on DownloadStats.report() to do it, but we don't
guarantee that will happen anymore. And really, when
Downloader.update_cache or Downloader.download_for_time finish, it's
done and should be marked as such.

It's no longer needed and it was an over-clever hack. #6

instead of magical logging.progress level. #6

DownloadStatistics.all_files() returns files that were downloaded or found
in the cache

Part of #5

Part of #5

To do this, DownloadStatistics is now mandatory and tracks files found.

This fixes the OTHER problem: the test files to download were too small
and could be plausibly retrieved within 1 second, so tests inside the
SSEC could fail to trigger the timeout.

Individual downloads need individual destinations. I don't know why it
was working before. I'm not sure why it wasn't working now.

It's still broken, but for a different reason! :-)

Eliminates the need for the scripts change their own sys.path. Also gets
the non-stock dateutil.parser

Slightly simplifies the import and testing.

Either should work, but the needed files disappeared from geodb.

If all of the filesets are NOT expected, it's okay, just log at INFO.

If any fileset IS expected, that's worrying and generates a WARNING.
ERROR is too severe because we might be able to continue, and there is a
good chance that the error is entirely remote.

3 hour delays are typical.
4.5 hours delays are moderately common.

Matches original aitf-mirror behavior in cache update mode.

It's purely speculative work: "Guess what people might find useful."

If we're --cache-only, we DEFINITELY need to use the user provided cache
directory, even if not writable.

Remove timestamp if period >= 1 day

We no longer emit the timestamp when self.period >= 1 day

Should be: 2020-01-01 12:00:00
Was incorrect: 2020-01-0112:00:00

aitf-data-for-run now uses "--cache PATH" while
aitf-update-cache continues to use a bare "PATH"

aitf relies on the behavior

Useful for GFS, which can fallback to a day earlier, but needs the 12+
hour forecasts, which we normally wouldn't get.