Fix CondaBuild CI job

[xmnlab]

Currently condabuild ci job is failing because:

• the current conda recipe doesn't work for python 3.8, because pymapd depends on pyarrow <= 0.13 and it has dependency conflict with another packages. also the current pyspark will not work well for python 3.8 (waiting pyspark v3)
• as conda recipe uses noarch:python, and currently the conda build uses the python 3.8 to build, the build will have a very complex dependency tree to resolve

in this PR:

• I changed the repo and the branch used to get the conda recipe to check if the new recipe will work
• I added the python flag to feedstock.py test to specify the python version to conda-build

notes:

• as can be observed, it worked. the time spent is not the desired yet (43m) but as the time limit is 1h, probably it is enough for now.
• would be good to improve more this, maybe in a follow up PR. something that could help to improve this would be: 1) use mamba for building and 2) change the pinning of some packages (in the conda recipe).

[jreback]

ok I merged the feedstock PR. do we need this one any more?

[xmnlab]

the time necessary for building the package on conda-forge using the current recipe:

https://dev.azure.com/conda-forge/feedstock-builds/_build/results?buildId=157972&view=results

after some pinnings:

https://dev.azure.com/conda-forge/feedstock-builds/_build/results?buildId=158182&view=results

the job for python 3.7 (on conda-forge feedstock) reduced from 1h22m to 0h07m

WIP PR: conda-forge/ibis-framework-feedstock#43

[jreback]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jreback]

thanks @xmnlab and @datapythonista

looks like we are back to green!

now to merge anything omnisci or spark we have to add back builds.