Resolve/document dependency discrepancies

[pont-us]

conda-forge releases (e.g. here ) trigger a dependency analysis which always finds a significant number of discrepancies between the dependencies listed in meta.yaml. Automated dependency anaylsis isn't 100% reliable, but we should resolve as many of these as possible and document the rest so we don't waste time looking into them again. for the 1.4.1 release, the list is as follows:

Packages found by source code inspection but not in the meta.yaml:

• cftime
• numcodecs
• botocore

Packages found in the meta.yaml but not found by source code inspection:

• werkzeug
• scipy
• adlfs
• python-blosc
• fontconfig
• lz4
• requests-oauthlib

Package	Comments	Action
cftime	imported in select.py etc.	add to environment.yml and meta.yaml
numcodecs	imported in zarrstore/generic.py	add to environment.yml and meta.yaml
botocore	imported in dsio.py	add to environment.yml and meta.yaml
werkzeug	only imported in tests	remove from meta.yaml
scipy	not imported	remove from environment.yml and meta.yaml
adlfs	not imported, but has comment ‘for azure blob filesystem’	clarify
python-blosc	not imported, but has comment ‘required by Coiled’	clarify -- see below
fontconfig	not imported	remove from environment.yml and meta.yaml
lz4	not imported, but has comment ‘required by Coiled’	clarify -- see below
requests-oauthlib	not imported	remove from environment.yml and meta.yaml

Notes:

• We'll also need to update DEFAULT_DEPENDENCY_NAMES in versions.py.
• From the above list, it appears that we specify explicit dependencies on two transitive (soft?) dependencies of the coiled package, but the coiled package is only a soft dependency (i.e. we try to import locally at runtime if/when it's actually needed and raise an error if it's not available). If these are hard dependencies of coiled, we can probably omit them entirely: if coiled is importable, they must be present. If they're soft dependencies of coiled which we require for our use case, we should handle them the same way we handle the coiled package itself -- i.e. check at runtime if/when required.

[pont-us]

Updates:

• After investigation and discussion, it seems fairly definite that python-blosc and lz4 aren't needed any more (if they ever were), so they should be removed entirely from the dependencies.
• python-blosc was the only package being installed from a non-conda-forge channel (since the Python-3.12-compatible conda-forge build of python-blosc is currently, and hopefully temporarily, broken), so having removed it we can also remove the defaults channel from environment.yml.
• adlfs seems to be a soft transitive dependency required only for Azure blob support. We should document:
  • which xcube dependency is optionally using it (currently unknown)
  • the fact that it needs to be explicitly installed if Azure blob support is needed
  • any other optional/soft dependencies which affect xcube's behaviour