xsar 0.6 pre release

[oarcher]

Bug fixes and improvements.

The main one is the use of dask Actor for Sentinel1Meta, so when running on a cluster, the s1meta instance is running on one worker, and might be serialized to other workers, without loosing it's state.

This induced a memory leak, because the Actor is never released. (ref https://github.com/dask/distributed/issues/5610)

As a temporary workaround, variable cache was disabled in xsar.xml_parser.py. (ref #23 )

[oarcher]

@alevieux , checks are in error state because some dependencies are missing ( see https://github.com/conda-forge/xsar-feedstock/pull/1)

[alevieux]

For update workflow GitHub action you can add rioxarray dependancies in environment install

for osx

https://github.com/umr-lops/xsar/blob/26ba7e98b32aa4301469af3d3103d88f6e080c4e/.github/workflows/mac-os-install-test.yml#L33

for ubuntu

https://github.com/umr-lops/xsar/blob/26ba7e98b32aa4301469af3d3103d88f6e080c4e/.github/workflows/ubuntu-install-test.yml#L33

for windows

https://github.com/umr-lops/xsar/blob/26ba7e98b32aa4301469af3d3103d88f6e080c4e/.github/workflows/windows-install-test.yml#L33

[alevieux]

We can delete the python 3.6 ubuntu test because the new dependencies rioxarray not support python 3.6 https://github.com/corteva/rioxarray/issues/215

In consequence xsar it's available with python>=3.7