Fix SDFITS index file writing to match sparrow format

[tchamberlin]

write_index() produced .index files that were incompatible with GBTIDL because the column names, header padding, format widths, and derived column computations all diverged from the sparrow3 reference implementation (IndexWriter.py).

What changed

Format fixes (verified character-for-character against sparrow3 .index.expected files):

• Column header row is now generated dynamically from sparrow3's format specs
• names like EXTENSION and POLARIZATION are truncated to EXT/POL
• Header line padding changed from 200 to 256 chars (based on sparrow code)
• Row format strings match sparrow3 exactly
• DATEOBS/TIMESTAMP width corrected from 21 to 22 chars
• Skip-spacing logic for #INDEX# and ROW columns at the 1M boundary (not sure what this is, but there's code for it in sparrow...)

Derived column stuff (port sparrow3's translateInfo):

• CENTFREQ is now computed from WCS parameters ((centerChan - CRPIX1) * CDELT1 + CRVAL1) instead of using CRVAL1 directly
• POLARIZATION derived from CRVAL4 (I/Q/U/V/RR/LL/XX/YY/etc.)
• PROCEDURE extracted from OBSMODE
• FREQINT from CDELT1, BANDWIDTH from BANDWID, SUBREF from SUBREF_STATE
• NSAVE defaults to -1 (was 0)
• SIG/CAL properly converted to T/F strings from booleans

No changes to the read path at all.

Tests

• The previously-skipped roundtrip test (test_write_and_read_roundtrip) is now passing
• Made some dysh-specific tests
• Ported 3 sparrow3 IndexWriterTests that comparing data rows character-for-character against *.index.expected reference

[mpound]

I like what you have done here. A few comments:

• Your solution for multi-file indices writes a single index file with different "FILE" values for each of the FITS files. This is different than GBTIDL which writes one index per FITS file, e.g. A.index, B.index, etc.
• The missing piece for dysh capability is that GBTFITSLoad/SDFITSLoad.write() would need a write_index boolean to invoke this. This would require a loop over write_index() passing in the SDFITSLoad.index for each FITS file. And test code to make sure it works correctly. At that point I'd also change dysh multifits write to use A, B, C instead of 0, 1, 2, 3...a little more bookkeeping.
• I'm inclined to accept this to get it into 1.0.0 so the base capability is there and then between now and panel review, implement sdfits.write(index_file=T/F)

[tchamberlin]

I think we are now at feature parity with SDFITS and GBTIDL's FITS writing capabilities, with a few exceptions:

1. GBTIDL seems to have some sort of interactive index writer. fileout/keep stuff
2. GBTIDL has some sort of spectra tagging system
3. SDFITS has a daemon that updates the index files in online mode; we don't try to duplicate that here. There's also some concept of lock files that doesn't seem necessary here, but not 100% sure

As usual I don't have time to do a full deep dive into this code, but we do have reg tests here against GBTIDL and SDFITS

[astrofle]

The default behavior should be to write the index file. However, changing the default to write index files also breaks a number of other tests (I have only checked the tests in the fits submodule):

• src/dysh/fits/tests/test_gbtfitsload.py::TestGBTFITSLoad::test_construct_integration_number
• src/dysh/fits/tests/test_gbtfitsload.py::TestGBTFITSLoad::test_write_all
• src/dysh/fits/tests/test_gbtfitsload.py::TestGBTFITSLoad::test_write_repeated_scans
• src/dysh/fits/tests/test_gbtfitsload.py::TestGBTFITSLoad::test_azel_coords
• src/dysh/fits/tests/test_gbtfitsload.py::TestGBTFITSLoad::test_hadec_coords
• src/dysh/fits/tests/test_gbtfitsload.py::TestGBTFITSLoad::test_galactic_coords
• src/dysh/fits/tests/test_gbtfitsload.py::TestWriteIndex::test_write_index_default_false

I can try to fix these when I find the time.