6fab403f removed the last requirement on glue and removed that package from the debian/rhel packaging, but not setup.py; this commit does that.

- following the migration of the document parsing code from array.py, param.py, table.py and lsctables.py into ligolw.py proper, it is no longer necessary to create application-specific content handlers.  the kind of customization that was being done in the past is now impossible.
- therefore, load_fileobj() can now assume a sensible default content handler
- all programs are updated accordingly, example code in documentation is updated accordingly
- because the APIs have now drifted too far, it is no longer possible to continue using the glue package to provide backwards compatible support for documents containing ilwd:char based tabular data.  this (now broken) feature is removed from the programs that were still providing it, and glue is removed from the package dependency metadata.

- move the lsctables.TableByName look-up table mapping table name to custom class from the module to a class attribute of ligolw.Table
- move the plumbing used to select and instantiate the appropriate class from the lsctables content handler to the Table.__new__() method
- this obsoletes the lsctables content handler.  now, it is sufficient to import the module to "enable" it.  replace the content handler with a deprecated compatibility stub
- move the dbtables.TableByName look-up table from the module to a class attribute of DBTable, overriding the new TableByName mapping in the parent class
- move the plumbing used to select and instantiate the appropriate class from the dbtables content handler to the DBTable.__new__() method

- convert to class method
- leave deprecated compatibility stub in its place

- also set the "valid" segments
- if "valid" segments are left blank, as the code was doing, then the new .coalesce() implementation that also merges lists deletes all the segments.

- .update() needs the amount to increment the progress by, not the value to set the progress to.

- for all lists that share the same metadata, merge them by applying .update() pair-wise, reducing them to a single list.

- python 2 no longer supported, packages are not built

- replace occurrences of this logic throughout the project with calls to this method.

- rearrange the code and add new section headings to help organize the module's contents.

- make a few aesthetic tweaks to the code

- instead of in-line logic to determine if the url names a local file, and if it does what the name is.

- reduce line count by removing xmldoc local variable

- Python now prints the attribute name in the error message when attempting to assign to a read-only attribute
- fixes borken doctest
- also clean up examples that show Tracebacks, replacing guts with "..." to reduce confusion over what the problem in a Traceback message is when doctest starts failing.

- convert to class method of Table class, inherited by classes in lsctables.

- convert to .register_process() method of Document() class

- inverse conversions are not supported at this time.  for string types there is no simple look-up table approach that is appropriate, the type string needs to be parsed.
- this change facilitates transforming tabular data to numpy arrays, and from there to other document formats, e.g. hdf5

- add more detail explaining how FromPyType look-up table works
- update dates in copyright notice

- access sys.stdin using python 3's .buffer only, removing the python 2 compatibility option

- now support only the ","-delimited format.
- supporting the legacy formats is no longer worth the cost.  there are no known users of the other formats.
- decoding and encoding instrument name sets was 10% of the cost of gstlal inspiral's coincidence code.  this patch reduces it to 2%.  still a problem, but also still a big improvement.

Kipp Cannon authored Jan 17, 2022 and Kipp Cannon committed Jul 21, 2022

- add ligolw_segments run test to test suite to catch problems like this in the future.

- not added to test suite because you use it by letting it run for hours to see if its ram footprint climbs.

- use an O(N) instead of O(N^2) algorithm for removing escape characters
- add a fast-path early out for strings with no escaped characters

- why one of the memory copies should not be considered costly

- performance improvement
- forgot that next_token() null-terminates the token, so floats and ints can be parsed in-place.  no need to copy to separate buffers.