=====================
* Rewrite the COARDS netCDF output to create xarray objects rather than netCDF datasets directly.
In some quick testing, this results in about a 3x speed up, this depends more on variable count vs data length, so most of the performance increase is actually in the bottle output
* Fixed a bug in COARDS where the fill value was not being set in the bottom depth variable
* Add `fill_values` and `precision_source` arguments to `read_csv`
* Add string literal types for the `ftype` parameter of `read_csv`
* CLI improvements:
* made "precision_source" and option rather than positional argument
* added a `--comments` option to allow the override of comments from either a string or file path prefixed with .
* Add a convert_csv subcommand which takes an additional ftype option to specify (C)TD or (B)ottle
* Removed the `matlab` optional install extra, this previously had a single dependency of "scipy" in it.
Scipy is used by xarray for netCDF3 output so this dependency has been moved to the `netcdf` optional install extra.