2.0.1

Three fixes, all in the same family: parameters and fit models the command
line advertises but that did not reach, or could not survive, the fit (#51).

Fixed

• --fit-model scipy-median-filter could not run in the image plane at
  all. FITS is big-endian by definition and scipy.ndimage.median_filter
  accepts only native byte order, so on every little-endian machine the fitter
  reached scipy with a >f4 spectrum and got back a bare RuntimeError: Unsupported array type, naming neither the array nor the reason. The
  exception escapes ContSub.fitContinuum, so it took the whole run with it.
  The spectrum is converted at that one call site now. The visibility plane was
  never affected: dask-ms yields native arrays.

• --chan-width is honoured rather than accepted and ignored. Both
  im-mowjsub and vis-mowjsub declared it and neither passed it to a fitter,
  so it satisfied no validation check and changed no result: a run given only
  --chan-width was refused for want of --vel-width, and a run given both
  quietly used the velocity. Giving both widths is now an error rather than a
  silent preference for one, and a --chan-width below 1 is refused.

Changed

• im-mowjsub --sigma-clip takes a single value, and --automask-per-iter
  is gone. Both described an iterative automasking mode that was never
  implemented, and the list type was unusable rather than merely unused —
  PixSigmaClip multiplies the whole list against the noise array in one
  operation, so any count but one mis-broadcast. A command line passing more
  than one --sigma-clip, or --automask-per-iter, must drop them; nothing
  that worked before stops working, since one value is all that ever did. The
  multi-iteration example in the usage docs goes with them.

Full changelog: https://github.com/laduma-dev/mowjsub/blob/2.0.1/CHANGES.md