Clearer, instead of accessing global variable d.

No need since we want all the variables in the file.

Bug fix: take into account ghost eddies in global Matlab adt
files. Some eddies are duplicated in global adt files, at the border
of the longitude domain. We should not use these ghost eddies. So
R\'emi has added variables `id_anti_cell` and `id_cyclo_cell` in adt
files. Read them. They should contain numbers from 1 to the number of
eddies, with a step of 1, intersparsed with NaN values. Check that
when reading the variables. Add an argument `cell_mask` to function
`write_to_shp` that signals ghost eddies.

Bug fix: add default values for arguments of function
`write_to_shp`. This fixes the case when the adt file is missing. The
bug comes from commit 28afccf5. For simplicity, we remove the empty
list actual argument for cell, in the case of a missing file. Then we
have to test in `write_to_shp` whether cell is None.

Move argument cyclone of function `write_to_shp` before argument
cell. cyclone does not depend on `matlab_data`. We are preparing a bug
fix for the case of a missing adt file.

The name write is too common.

Instead of using values of global variable `matlab_data` inside the
function write. This is clearer. Especially since we already had an
argument of write, cell, which is a value of `matlab_data`.

Required since commit ed60f104.

Much simpler and more robust. This should also ease submitting the
script to a batch scheduler. And we can now check the return code of
Matlab.

Take advantage of version 2019 of Matlab now available on Ciclad,
which has option `-batch`. This clears the output and opens the way to
subprocess instead of `pty.spawn`.

So we do not risk loading `adt_v6.mat` from a previous iteration if
Matlab fails in the current iteration.

This is useful when we run the script through batch scheduling.

It is not enough to parse the input file name and extract the date
because, when we process other dates, we have to construct file names
in the same format. So do not use `parser.parse` for the input file
name, try various formats and remember the one that works.

Method poly of shapefile writer requires a list of lists, not a list
of numpy arrays. This bug goes unnoticed in pyshp `2.1.0` but raises an
exception with pyshp `2.1.3`.

Revert part of commit dfb78dbc. Motivation: being able to use
`plot_eddy_contours.py` on results from Matlab.

We follow Matlab naming. We want the transition in output format to be
as easy as possible.

Rename field `date_index` of shapefiles to `days_1950`. Clearer.

We just write a line in `ishape_last.txt`. Note that we cannot catch
the error, which occurs inside Matlab, so we have to test for file
existence beforehand.

Files for EUREC4A have format `adt_YYYY-MM-DD.mat` while files for
domain PhD-Lax and global domain have format `adt_YYYY-MM-DD.mat`, so we
need something more general than strptime.

Write to `SHPC_(anti|cyclo)` instead of `Snapshot_(anti|cyclo)` since
we are now writing several dates.

Call the Matlab script from the Python script. We cannot do this with
subprocess because Matlab requires a terminal. With subprocess, the
Matlab script complains: "inappropriate ioctl for device". Note that
we cannot test that the Matlab script worked normally because it exit
with code 0 regardless: the exit instruction returns code 0.

The idea here is that we do not want to store the converted v6 mat files.

We already have it: it is `sys.argv[1]`.