BANDSB rework

[bradh]

This updates the existing BANDSB implementation to parse representative sample data. It extracts data from the same sample used for MATESA, and builds on #497 infrastructure.

The key changes are to the TRE parsing, because BANDSB uses a four-byte float representation (i.e. IEEE-754 binary) as well as the more typical ASCII encoded floating point values (and hyphen-minus fill versions of that). I've added 'f' for the ASCII floating point representation, and 'F754' for the IEEE-754 binary representation. A better name would be appreciated.

[pressler-vsc]

Thanks for this PR @bradh; especially for including the test data and test. We had a few concerns (and found another underlying SarPy bug) for which we've submitted a slightly different approach in #502. Does the modified behavior in that PR still address your use case?

[bradh]

Haven't tried it, but I don't see any problems. Will close this in favour of #502.