MRG, MAINT: Improve server env

[larsoner]

I really just want to add nibabel and nilearn to our server_environment because they are needed for some viz functions, but then I noticed some other stuff, so this PR will:

1. Add nibabel and nilearn
2. Remove vtk_dev channel
3. Remove defaults channel
4. Remove the numpy<1.19.0 pin

Hopefully it all works but if it doesn't, hopefully I can at least get (1) in there. I'll need to double-check that the env otherwise looks the same in terms of what gets installed by looking at mne sys_info.

[larsoner]

Old:

Platform:      Linux-5.4.0-1031-azure-x86_64-with-glibc2.10
Python:        3.8.5 (default, Sep  4 2020, 07:30:14)  [GCC 7.3.0]
Executable:    /home/vsts/miniconda/bin/python3.8
CPU:           x86_64: 2 cores
Memory:        Unavailable (requires "psutil" package)
mne:           0.22.dev0
numpy:         1.18.5 {blas=openblas, lapack=openblas}
scipy:         1.5.4
matplotlib:    3.3.3 {backend=agg}

sklearn:       Not found
numba:         Not found
nibabel:       Not found
cupy:          Not found
pandas:        Not found
dipy:          Not found
mayavi:        Not found
pyvista:       0.27.4 {OpenGL 3.3 (Core Profile) Mesa 18.3.1 via llvmpipe (LLVM 7.0, 256 bits)}
vtk:           9.0.0
PyQt5:         Not found

New:

Platform:      Linux-5.4.0-1031-azure-x86_64-with-glibc2.10
Python:        3.8.5 (default, Sep  4 2020, 07:30:14)  [GCC 7.3.0]
Executable:    /home/vsts/miniconda/bin/python3.8
CPU:           x86_64: 2 cores
Memory:        Unavailable (requires "psutil" package)
mne:           0.22.dev0
numpy:         1.19.4 {blas=NO_ATLAS_INFO, lapack=lapack}
scipy:         1.5.3
matplotlib:    3.3.3 {backend=agg}

sklearn:       0.23.2
numba:         Not found
nibabel:       3.2.1
nilearn:       0.7.0
dipy:          Not found
cupy:          Not found
pandas:        1.1.4
mayavi:        Not found
pyvista:       0.27.4 {OpenGL 3.3 (Core Profile) Mesa 18.3.1 via llvmpipe (LLVM 7.0, 256 bits)}
vtk:           9.0.0
PyQt5:         Not found

@GuillaumeFavelier I got a segfault when I tried to remove vtk_dev -- any idea why we have to use that version rather than the one provided by conda-forge?

[larsoner]

mne and jupyter are stuck in pip because they require / pull in PyQt5, which often breaks headless servers (anything that tries to import PyQt5 will die because Qt5 throws an error)

[GuillaumeFavelier]

any idea why we have to use that version rather than the one provided by conda-forge?

To fetch the version of vtk built with osmesa. At least, it was the case when I added this conf.

[larsoner]

Pushed a commit for the yield_fixture business, too.

@GuillaumeFavelier you happy? If so I'll merge when green

[GuillaumeFavelier]

LGTM