This implements an indexer that provides the methods necessary to
shard a catalog and then look up the correct shard later.  The package
I use currently has a problem if you call the lookup_id method without
first calling another function.  This hack will go away if the package
is fixed or if we get an LSST maintained version.

This will be used to read files from disk for indexing

The catalog ingester will load files using a helper task
and put the contents in a set of SourceCatalogs, one
per HTM trixel.  Most of the config is to allow mapping of
columns in the file to the schema needed for referenc catalogs.

This implements the task to take files and parse them into
SourceCatalogs.  Some of the tasks implemented here are:
read the file into a numpy array (the whole file needs to fit
in memory), construct the SourceCatalog schema from the array
dtype, index the catalog and fill the SourceCatalog.

Command line tasks need a command runner.  Since this task
doesn't need to parse an `--id` argument, we an forego that
step.  Essentially, this just calls the task run method and
gives a place to persist the config.

This involves reconstituting the task that ingested the files.

The only config is the name of the config to read to create the
ingester task.  The loader then borrows the indexer from the
ingester task to grab the files from the repo it needs to load the
reference catalog.  The shards are only clipped if they land on the
boundary of the circular aperture.

This is a minimal override of the reference catalog loader.  It may
be possible to optimize by overriding other default implementations.