Shim class for loading G3SuperTimestream objects from disk

[arahlin]

This PR is the first step in integrating the G3SuperTimestream class used by the Simons Observatory team into the core library. This class is a very thin wrapper around the core G3TimestreamMap class. It is designed only to load G3SuperTimestream objects from G3 files on disk for immediate conversion to G3TimestreamMap objects. It is also not exposed to the public or python API.

[arahlin]

A couple of missing features that I've noticed so far:

• There's no check that the times array is contiguous and monotonic. We should probably have some code that fills in the missing samples with nans. EDIT: add a check for missing time samples, gap-fill with zeros and add a bonus detector channel with the injected samples flagged
• There's also no check that the names array is ordered, which G3TimestreamMap requires. Probably the right thing to do is to populate the map in an ordered way (carefully tracking offsets and quanta, etc). EDIT: changed G3TimestreamMap backend to preserve insertion order instead.
• The implementation here decompresses all of the data in memory immediately on load, rather than storing the compressed blob for later access. Changing this behavior is maybe feasible within the G3TimestreamMap class, but depends on whether its actually used / useful in practice.
• The G3TimestreamMap class currently only uses flac serialization for a particular kind of data (24-bit ints, in units of Counts). Compression of the full 32- or 64-bits is straightforward to implement as of flac version 1.4. EDIT: added an option to compress / decompress int32 timestreams at full depth.
• The type_num item stores a pretty fragile numpy enum whose possible values appear to be platform dependent. I hardcoded these values using what is likely the most commonly used set of them, since this code isn't designed to fix that sort of issue. However, it's possible that old data created on some system that defines these numpy types differently will not be readable with this code. EDIT: handled these the same way that G3Timestream handles them with the __LP64__ preprocessor directive.

[tskisner]

Thanks for this work! For the record, I tested this branch in combination with a draft PR of so3g that allows building against newer spt3g versions. With that combination, I was able to load some Simons Observatory data through this shim into the downstream sotodlib and toast data containers.

[arahlin]

I think this is ready to merge, and we can iterate on any missing features in G3TimestreamMap separately. Any objections, @nwhitehorn or @mhasself ?