Use environment.yml (or something like that) in CI

[ricardoV94]

Description

The CI dependencies are completely dissociated from the conda pytensor-dev environment specified by environment.yml.

This led to need a separate commit in #448 that should have gone into #446

pytensor/.github/workflows/test.yml

Lines 139 to 152 in a970407

	- name: Install dependencies
	shell: bash -l {0}
	run: |
	mamba install --yes -q "python~=${PYTHON_VERSION}=*_cpython" mkl numpy scipy pip mkl-service graphviz cython pytest coverage pytest-cov pytest-benchmark pytest-mock sympy
	# numba-scipy downgrades the installed scipy to 1.7.3 in Python 3.9, but
	# not numpy, even though scipy 1.7 requires numpy<1.23. When installing
	# PyTensor next, pip installs a lower version of numpy via the PyPI.
	if [[ $INSTALL_NUMBA == "1" ]] && [[ $PYTHON_VERSION == "3.9" ]]; then mamba install --yes -q -c conda-forge "python~=${PYTHON_VERSION}=*_cpython" "numpy<1.23" "numba>=0.57" numba-scipy; fi
	if [[ $INSTALL_NUMBA == "1" ]] && [[ $PYTHON_VERSION != "3.9" ]]; then mamba install --yes -q -c conda-forge "python~=${PYTHON_VERSION}=*_cpython" "numba>=0.57" numba-scipy; fi
	if [[ $INSTALL_JAX == "1" ]]; then mamba install --yes -q -c conda-forge "python~=${PYTHON_VERSION}=*_cpython" jax jaxlib numpyro && pip install tensorflow-probability; fi
	pip install -e ./
	mamba list && pip freeze
	python -c 'import pytensor; print(pytensor.config.__str__(print_doc=False))'
	python -c 'import pytensor; assert pytensor.config.blas__ldflags != "", "Blas flags are empty"'

[lmmx]

The only tricky bit here is that you can't parameterise the Python version of an environment.yml, but we can do the installation of the parameterised Python version and the environments file in the same pass using this pattern