* Add python 3.8 support and deprecate python 3.5

This commit adds support for running aer under python 3.8. It does
this by adding the trove classifiers to the package metadata and adding
a test job for python 3.8  At the same time this commit starts the
deprecation window for python 3.5. Python 3.5 goes end of life by
upstream python in Sept. 2020. We should give our users running with
python 3.5 (which is about 10% of qiskit users this year based on pypi
data) fair notice that when upstream python stops supporting it we do as
well. Corresponding to this change, terra is going to start raising a
warning when anything from qiskit.* is imported, see:
Qiskit/qiskit#3268

* Install numpy and scipy from conda forge

For whatever reason the scipy package on the windows python 3.8 jobs is
trying to build from sdist. This requires blas and lapack libraries
installed because it is building the package from source. Instead of
relying on pip to install scipy this just switches to using the
conda-forge version of numpy and scipy which are precompiled for windows
and should work fine.

* Add conda jupyter package to 38 windows job

* Conda install pywin32 too

* Cap jupyter-sphinx version to non python 3.8 jobs

The python 3.8 windows ci job can not install jupyter sphinx because of
an issue with the pywin32 package. Pip is unable to find a compatible
package for this and conda installing it doesn't satisfy the requirement
for pip either. To workaround this issue this adds a version cap on
jupyter-sphinx so we do not try to install it by default in every test
environment. The package is only needed for building documentation and
since it's unlikely that people will be installing docs on python 3.8 at
this point and our ci jobs do not we use this to unblock python 3.8 jobs
on windows temporarily. In the future though we'll have to remove this
when people start using 3.8 as their default interpreter so that
building docs with python 3.8 is feasible.

* Stop trying to rely on conda-forge and just try pip

* Add -v to pip install for debug

* Don't upgrade pip and friends for python 3.8 windows job

* Add osx job to premerge ci too

* Fix typos in travis config

* Install osqp via conda

* Fix typo in changelog

* Switch to bash for conda install on 3.8

The script directive doesn't look like it's every command specified. The
output for the python 3.8 anaconda install step doesn't seem to include
the output for the second conda install command. This commit switches it
to use bash so we can set trace mode to see if we're even running the
command, or if something else is happening.

* Fix variable access

* Break out 3.8 requirement binary stage and switch to cvxpy

osqp is only installed as a dependency for cvxpy. So instead of trying
to install osqp and getting pip to see that the dep is already
satisified when installing cvxpy it'll be easier to just install cvxpy
via anaconda instead.

* Back to osqp cvxpy fails to install

* Tweak stages on azure pipelines slightly

The azure pipelines stages had unecessary steps and duplicated logic.
This commit cleans it up slightly and makes it easier to follow.

* Try splitting out dependency installs

Oddly it looks like pip isn't finding osqp when installing the
requirements outside in a separate stage. This commit attempts to fix
the issue by moving the dependency installation into it's own stage.

* Install more python deps via conda

* Fix bash variable access again

* Add python 3.8 support and deprecate python 3.5

This commit adds support for running aer under python 3.8. It does
this by adding the trove classifiers to the package metadata and adding
a test job for python 3.8  At the same time this commit starts the
deprecation window for python 3.5. Python 3.5 goes end of life by
upstream python in Sept. 2020. We should give our users running with
python 3.5 (which is about 10% of qiskit users this year based on pypi
data) fair notice that when upstream python stops supporting it we do as
well. Corresponding to this change, terra is going to start raising a
warning when anything from qiskit.* is imported, see:
Qiskit/qiskit#3268

* Install numpy and scipy from conda forge

For whatever reason the scipy package on the windows python 3.8 jobs is
trying to build from sdist. This requires blas and lapack libraries
installed because it is building the package from source. Instead of
relying on pip to install scipy this just switches to using the
conda-forge version of numpy and scipy which are precompiled for windows
and should work fine.

* Add conda jupyter package to 38 windows job

* Conda install pywin32 too

* Cap jupyter-sphinx version to non python 3.8 jobs

The python 3.8 windows ci job can not install jupyter sphinx because of
an issue with the pywin32 package. Pip is unable to find a compatible
package for this and conda installing it doesn't satisfy the requirement
for pip either. To workaround this issue this adds a version cap on
jupyter-sphinx so we do not try to install it by default in every test
environment. The package is only needed for building documentation and
since it's unlikely that people will be installing docs on python 3.8 at
this point and our ci jobs do not we use this to unblock python 3.8 jobs
on windows temporarily. In the future though we'll have to remove this
when people start using 3.8 as their default interpreter so that
building docs with python 3.8 is feasible.

* Stop trying to rely on conda-forge and just try pip

* Add -v to pip install for debug

* Don't upgrade pip and friends for python 3.8 windows job

* Add osx job to premerge ci too

* Fix typos in travis config

* Install osqp via conda

* Fix typo in changelog

* Switch to bash for conda install on 3.8

The script directive doesn't look like it's every command specified. The
output for the python 3.8 anaconda install step doesn't seem to include
the output for the second conda install command. This commit switches it
to use bash so we can set trace mode to see if we're even running the
command, or if something else is happening.

* Fix variable access

* Break out 3.8 requirement binary stage and switch to cvxpy

osqp is only installed as a dependency for cvxpy. So instead of trying
to install osqp and getting pip to see that the dep is already
satisified when installing cvxpy it'll be easier to just install cvxpy
via anaconda instead.

* Back to osqp cvxpy fails to install

* Tweak stages on azure pipelines slightly

The azure pipelines stages had unecessary steps and duplicated logic.
This commit cleans it up slightly and makes it easier to follow.

* Try splitting out dependency installs

Oddly it looks like pip isn't finding osqp when installing the
requirements outside in a separate stage. This commit attempts to fix
the issue by moving the dependency installation into it's own stage.

* Install more python deps via conda

* Fix bash variable access again