Avoid timeouts / storage exceeded in PR tests

[dagewa]

We are frequently seeing failures in various GitHub actions running on pull requests. Some of these might be throttling causing download failures of dials-data data sets, some might be due to disk space being exceeded. What can we do to improve this situation?

Related

• Compress / remove / edit legacy image formats to reduce the size of downloads (Do we need to test support for legacy formats? cctbx/dxtbx#857)
• Compress .refl files to reduce the size of downloads (Compress reflection files data-files#69)
• Move the location of some files elsewhere (a server at LBL has been mentioned)