More informative error message for error reading InferenceData from NetCDF

[rpgoldman]

Description

Previously, when trying to read an InferenceData object using from_netcdf on a filesystem that did not support HDF5 file locking (full disclosure: I have no idea what HDF5 file locking is), one would get a cryptic OSError: -101 which did not provide any assistance in fixing the problem.

I tracked this down, and found a solution on StackOverflow. Because it's difficult to figure out (happens due to a dependency that is well down in the call tree), I tweaked from_netcdf to trap this error, and add an explanation.

Checklist

• Follows official PR format
• Code style correct (follows pylint and black guidelines)

Remarks

WRT the code style, running pylint locally I get it complaining that the use of e as a variable name is bad, but this is quite normal usage, stylistically, for variables that will be bound to Exception's.

[codecov]

Codecov Report

Merging #1637 (8aa810d) into main (23e14fb) will decrease coverage by 0.02%.
The diff coverage is 69.23%.

❗ Current head 8aa810d differs from pull request most recent head c7eba46. Consider uploading reports for the commit c7eba46 to get more accurate results

@@            Coverage Diff             @@
##             main    #1637      +/-   ##
==========================================
- Coverage   90.91%   90.89%   -0.03%     
==========================================
  Files         108      108              
  Lines       11671    11671              
==========================================
- Hits        10611    10608       -3     
- Misses       1060     1063       +3     

Impacted Files	Coverage Δ
arviz/data/inference_data.py	83.14% <69.23%> (-0.46%)	⬇️
arviz/plots/posteriorplot.py	91.17% <0.00%> (-0.50%)	⬇️
arviz/rcparams.py	93.47% <0.00%> (-0.49%)	⬇️
arviz/data/io_pystan.py	95.68% <0.00%> (-0.41%)	⬇️
arviz/data/io_cmdstanpy.py	96.12% <0.00%> (-0.38%)	⬇️
arviz/data/io_cmdstan.py	90.90% <0.00%> (-0.05%)	⬇️
arviz/stats/stats.py	96.67% <0.00%> (-0.01%)	⬇️
arviz/data/io_pymc3.py	90.75% <0.00%> (ø)
arviz/plots/distplot.py	90.47% <0.00%> (ø)
arviz/data/io_numpyro.py	94.48% <0.00%> (+0.59%)	⬆️
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 99764f6...c7eba46. Read the comment docs.

[OriolAbril]

LGTM, only needs to be added to the changelog

[OriolAbril]

To use e (instead of err or something a bit longer that pylint considers optimal) as variable to raise errors from other errors, it may be worth it to add it to .pylintrc instead of having to disable the error every time

[rpgoldman]

@OriolAbril @ColCarroll This is ready to go, if you are OK with it, please squash and merge.