Upgrade from Python 3.6 to 3.7

[weiji14]

Lots of new goodies in Python 3.7! Mainly nice to haves e.g. better parallelization with improved asyncio usability via asyncio.run(), easier debugging using breakpoint(), etc. Most of this DeepBedMap's dependencies have supported Python 3.7 for months already, all except onnx-chainer but that's changed with https://github.com/chainer/onnx-chainer/pull/192. Woohoo!

TODO:

• [x] Update onnx-chainer from 1.4.1 to 1.5.0 which support Python 3.7 (4fd1f84a04e3bc5691173adc18beb4a1f556abd2)
• [x] Use Python 3.7 in our environment.yml and Pipfile (c80041dcedec03f872225ce116b4c0c500b3d524).
• Try to update netcdf4 from 1.4.1 to 1.5.1.2, pinned since a84dfd6bce883bb075c0e874749b657a92a13b4c (see also #58)

[weiji14]

Could have updated netcdf4 from 1.4.1 to 1.5.1.2, but that would require compiling the netcdf4-python package from source. No easy way to specify that in the Pipfile, but we could use the PIP_NO_BINARY environment variable (see this Stack Overflow post) and do it like so:

PIP_NO_BINARY=netCDF4 pipenv install netcdf4==1.5.1.2

but that won't be accurately reflected in the Pipfile.lock (?) that would undermine reproducibility as compiling the package at different times (with different compilers) may yield slightly different netcdf4 library products :frowning_face:

Wait, but sticking to netcdf==1.4.1 and using python 3.7 still means we are compiling it (that's why we see this and also why $HDF5_DIR was set) :man_facepalming: , but at least we don't need to modify our Dockerfile to set a new $PIP_NO_BINARY environment variable... Anyways, too out of scope for this Pull Request, so pushing that forward to the future.