We really need to generalize the code that transfers one
file to another location so that ingest and export use
the same code and preferably can be shared amongst different
datastores. You can imagine ingest from S3 to local file or
export from S3 to local file or export from local file
to S3.  It is not scalable to put this code in multiple
places in multiple datastores.

Now you get different subclasses for schemeless, file
and generic URIs.  This simplifies some of the
logic for handling path components.

This now leaves us with a tiny bit of duplication
between schemeless and file fixups.

Defer converting to posix form until the end, thereby letting
us use os.sep everywhere.

ButlerURI.transfer_from(URI) now works for file, http, and S3.

It is a surprise to see a schemeless relative path become
a schemeless absolute path when asking for the directory
component. Changing this required a couple of tests to
be updated that were assuming absolute paths.

mypy gets confused that new does not return a ButlerURI
which leads to mypy thinking that we are calling
abstract methods.

Simple routine that special cases schemeless URIs.

Now unified resource access for files, S3, HTTP, and pkg_resources.
Much simplified I/O.

Still need to sort out dumpToFile

This helps in cases like tempfile.mkdtemp where the path
returned does not include a trailing slash but is used
as a string and not a ButlerURI.

This removes all the special casing for dumpToFile and
dumpToS3 in Config.

Removes a lot of code.

Absolute paths with a relative path should now work.

At some point we should go through and require that
datastore.root be a URI and not allow it to sort of
be a file.

Calling realpath on the destination for the transfer
is never the right thing to do. It was harmless when the
destination did not exist but in cases where the destination
existed and was a softlink that ended up somewhere else
it completely moved the output location.

on macOS the /var folder is a link to /private/var which
means that when you create a temp directory in a test
and put a file in it, once you readpath you end up
in a completely different location that is nowhere
near the test file that you are trying to use
with relsymlink.  Using a local tempdir fixes
this anomaly.

This enables posixDatastore to use it.

files containing ? were breaking because urllib.parse would
treat this as a URI with query parameters. We therefore need
to quote to protect this prior to parsing in some cases and
also unquote when referring to local file resources.

Rather than looking for a substring which is easy to miss
at a glance.

This required that we also add mem:// support for in-memory
datastores.