Adopt PEP 517 and PEP 621, use raven-hydro package

[Zeitsperre]

Closes #283

Changes

• Replaced setup.py with new format pyproject.toml using the flit backend.
• Updated pre-commit hooks to include formatters and checkers for TOML files.
• Removed the obsolete requirements*.txt and Manifest.in files.
• Dealt with an import-based error that occurred due to the sequence in which modules are loaded at import (attempting to call ravenpy before it is installed).
• The build recipes no longer build on each other, so when installing the dev or docs recipe, you must also install the gis recipe. A workaround I've added is to simply list the gis recipes in both for now
• This PR also updates the GeoServer API calls to at the very least work with the newest GeoPandas (using urllib). A more elegant solution should be looked into (to synchronize owslib / requests / urllib.request calls).
• This PR leverages the new method of installing the Raven model using https://github.com/Ouranosinc/raven-hydro.

What still needs to be addressed

• There is now no method of building raven-hydro from PyPI. The current approach was rendered obsolete in pip>=23.1, so we are in the right to retire this method.
  • A new method of installing raven-hydro using scikit-build-core is in the works.
    • This has been implemented (https://github.com/Ouranosinc/raven-hydro); A release on PyPI is forthcoming.
• The CI for the PyPI builds using tox are broken due to the missing raven-hydro binary. There is likely a whole slew of things that must be addressed in tox (removal of tons of now-obsolete workarounds), so don't expect this to work yet.
  • These issues have been resolved with the new installation method (https://github.com/Ouranosinc/raven-hydro); A release on PyPI is forthcoming.
• There are documentation files as well as the top-level Makefile that refer to operations using setup.py, and twine. These need to be replaced with their flit-based equivalents. For more information, see: https://flit.pypa.io/en/stable/index.html
  • These issues have been resolved.

Why?

With new standards for tooling and packaging, the setuptools-based compilation approach was very quickly becoming unmaintainable and only supported building on Linux/Unix platforms.

For raven-hydro, scikit-build-core is a setup backend that supports multi-OS builds using CMake (which is a much more approachable crossplatform C/C++ build engine). This renders the setuptools-based installation approach obsolete, as we are now able to install a Python wheel or source distribution specifically for the Raven model that allow users to compile it themselves on pip install (similar to numpy).

[Zeitsperre]

The new fiona-related errors are due to changes in geopandas that affect reading files from URLs: geopandas/geopandas#2796

The fix is to simply adjust the way we handle these objects in ravenpy.utilities.geoserver. Benefits include significant read speedups.

[huard]

Point to the Raven documentation ?

[huard]

Just make sure that the Install docs are clear on what needs to be done now, ie. use conda or compile yourself.

[huard]

Is this because there is deprecation warning?

[Zeitsperre]

This is because we have just changed it: Ouranosinc/xclim@0e06e2c

I think we need to pin this to the newest version of xclim as a result!

[Zeitsperre]

Either that or we can make it backwards-compatible with a try/except block.

[huard]

Ok, let's pin to the newest xclim then.

[Zeitsperre]

I think I've earned this.

[Zeitsperre]

There's an environment inconsistency for conda:

• The newest raven-hydro is built against the newest libnetcdf/libhdf5/etc.
• The last build of climpred on conda-forge was in November 2022 (no updates since)

As it stands, the libraries are incompatible:

  - climpred               2.1.6  pyhd8ed1ab_1                 conda-forge                    
  + climpred               2.3.0  pyhd8ed1ab_0                 conda-forge/noarch       Cached
  - raven-hydro            0.2.0  py310h3ce4ad4_0              conda-forge                    
  + raven-hydro              3.6  ha2b8ddb_1                   conda-forge/linux-64     Cached

I think climpred needs to be regenerated, or we need to offer pinned versions of netCDF (which I really very much don't want to do).

[Zeitsperre]

@huard

It's looking good! The last thing that needs to be revised is the one failing test on macOS. I think it has to do with the CRS not being read on file open. One approach would be to change the engine being used to read the file (try GeoPandas?) Will explore this tomorrow.