Add support for loading data from Amazon S3

[ericthewizard]

Heliocloud has a large amount of MMS data available via S3 buckets. At some point, we should update PyTplot and PySPEDAS to support loading data from S3

[jameswilburlewis]

Some notes from PyHC discussion on containerizing PyHC projects:

fsspec package can be used to open files on S3 and return an object that behaves as an ordinary file handle. It also supports caching to local filesystem.

cdflib already supports opening files on S3. netcdf4 does not, but there is an s3netcdf4 package that supports this.

It might be good for cdflib and netcdf4 to provide direct support for passing fsspec objects instead of filenames (abstracting out the type of endpoint being used).

[edmondb]

Created a working branch (416-s3-awareness) as we begin to address this. Let us know if you'd like to discuss any updates and/or requests associated with this.

[jameswilburlewis]

Looking forward to seeing what you come up with! Feel free to bounce any ideas off me...do you have a general approach in mind yet? I was thinking it would be nice if the CDF, NetCDF, and FITS readers could be modified to take open file descriptors instead of filenames, then if the PySPEDAS download code sees s3: URLs, it could open them itself (using an S3 aware package) and pass the handles to the other libraries.

[edmondb]

@jameswilburlewis , this update with associated PR is not a full solution to close this issue. It covers the majority of PySPEDAS cases that calls download.py but does not include dependencies (like fsspec), custom download functions, and further development will be coming in the future and pushed to this branch (documentation, tests, etc.). We could merge now with master (after we push the dependencies that are breaking the build tests) and then keep developing off of this branch if that's preferred. This push is to get the conversation going as to any additional comments on the PR.

Included here is also a flow chart of the logic for the download_file method within download.py. As you can see, it adds several functional decision paths to retain a non-API-breaking implementation for cloud-awareness.

flow.pdf

It would be great to receive feedback on this initial PR in order to ensure the remaining changes planned for the custom download methods will match in logic and functionality.