CI: Prefer conda-forge over defaults

[lithomas1]

• closes DEPS: pyarrow support for snappy not built on Windows Python 3.9 #42370
• tests added / passed
• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

Apparently its recommended to install pyarrow from conda forge instead of the default conda channel(xref ContinuumIO/anaconda-issues#12314, https://arrow.apache.org/docs/python/install.html#using-conda), trying to see if this fixes it. I don't think this will mess up any other packages since conda-forge packages are published by package maintainers, so they shouldn't have build issues?

[lithomas1]

Windows CI failure is a flaky ResourceWarning, feel free to restart the job(I cannot).

[jorisvandenbossche]

Thanks! Yes this should be fine.
The main reason that pyarrow mentions to install from conda-forge for the latest version is because of the default channel often being behind and not having the latest version built (that doesn't seem to be the case now, but so indeed not all features are included ..). I commented on the linked Anaconda issue about how this can be fixed on their side.

[simonjayhawkins]

ci failure on numpydev also also seen on other PR #42367

=========================== short test summary info ============================
FAILED pandas/tests/scalar/timedelta/test_arithmetic.py::TestTimedeltaMultiplicationDivision::test_td_div_nan[nan1]
= 1 failed, 156395 passed, 15976 skipped, 857 xfailed, 6 xpassed in 1944.90s (0:32:24) =

[simonjayhawkins]

Thanks @lithomas1