Cache generate_serialization_test_data.py's output to speed up CI

[notfilippo]

To speed up the CI times we should cache the sketch files in CI, keying them by commit hash of the repositories that generated them.

If datasketches-cpp or datasketches-java weren't updated in between CI runs there is no value in re-generating the serialised files so we should leverage the CI caching logic to make it more efficient.

[tisonkun]

You can follow https://lists.apache.org/thread/5ppjzrc151n387n4xfgpx1cxnhcv2sbk where I'm trying to use a shared repo to hold these snapshots so that we can simply download (or git clone) the files, and it would significantly reduce CI time.