Improve handling of MS file in filter_skymodel.py

This MR addresses issue #12 (closed) by removing the copy of the full MS in the filter_skymodel.py script and by resetting the TMPDIR env var after PyBDSF has run (where it may still be useful due to the socket paths length limits, though this may not be such an issue any more; see https://github.com/DataBiosphere/toil/issues/3380). Instead of the full copy, a minimal MS (with a single time slot and frequency channel) is generated instead. In tests, the attenuated flux densities generated by LSMTool from the minimal MS agree to better than 0.1% with those derived from the full MS.