Known bug with loading parameter files with rythmic inputs. Something
with displaying the spectrogram.

Prep for hnn-core integration by first reducing the number of times
a paramter file is opened. Instead pass the params dictionary where
convenient.

This commit replaces some calls to read functions in paramrw with
hnn_core.read_params, but some calls remain.

Parameter typing is done in hnn_core, use that when possible instead
of adding checks in the code (note paramrw.py functions).

Doesn't require passing a parameter anymore. Also, the number of data files
may not match the number of trials in the configuration (e.g. failure on
previous simulation)

Changes to run.py to call hnn_core.simulate_dipole using MPIBackend.

Updated hnn_qt5.py to not fork a mpiexec process, but instead call
simulatio() within run.py

Parameter handling:
 - remove read, quickreadprm, and quickgetparm from paramrw.py
 - use params['sim_prefix'] where possible

Simulation output files:
 - write spk, dpl, param, and gid_dict files after simulation
 - fix error handling in simdat.updatedat

Misc:
 - mark files and functions to be deprecated/moved
 - add excepthook to RunSimThread

For windows that are launched as a separate process, only read the param
file once with hnn-core and pass params as necessary. This avoids the
need for globals in most cases.

Globals still exist in visrast.py -- more work needed there

Also this completes using 'spec_cmap' with hnn core param files

Make these a list of dictionaries instead of a list of tuples

When the user clicks the button to run a simulation, the parameters from the
GUI are immediately saved to a file using the __str__ function of each Qt
dialog box. Then using hnn-core.read_params(), the dictionary of parameters
is returned. This dictionary is passed to RunSimThread for both regular
simulations and optimization batches.

The QComboBox functions are simplified and updated for the new format of
simdat.lsimdat

This is a large commit, with the major changes:

1. hnn is now a proper module. code is in the directory hnn, there is a setup.py,
   hnn/__init__.py, and relative imports are used
2. installation instructions were updated to complete the shift to pip install
   NEURON except for windows.
3. Travis CI is updated to install NEURON, but tests have not been migrated
   yet (they will fail)

Resolves jonescompneurolab#236

NEURON doesn't have a standard pip installation that will work
for Winodws yet. Until then, we cannot install hnn-core
using the default dependencies, which might pull in an incorrect
version of NEURON.

This makes the conda install two steps: 'conda create' and 'pip install'

Just start counting flake8 errors and set a goal for which files
will should be fixed. This returns zero (for now).

The dialog and GUI tests aren't working well. Try again once more work
has been done on the view*.py code.

Move around code that creates a new user with a space "test user"
to setup-travis-windows.sh run running py.test as a different user.

Renamed from run-pytest-wsl.sh

Also use params['sim_prefix'] instead of the QLineEdit text field

Also print a message to the console when the params can't be loaded

disable some tests on windows

This replaces the use of global variables for storing simulation
data.

simdat.py is now flake8 compliant

Large change that refactors plotting functions in simdat.py
to reuse code when plotting in the GUI and when creating a
plot to save to a .png.

Only two types of figures are saved: dpl and spec. Other figures
can be added in the future. Another limitation is displaying
poisson inputs along with evprox and evdist; it's not supported
right now.

The spectrograms are configured to cut of the first 50ms to
avoid edge effects. This produces not so visually pleasing figures
when saving. In the GUI, the dipole also loses the first 50ms!
A better way of choosing the time range for the spectograms needs
to be devices.

Send a signal with the results of a simulation back to the main thread
for storing in the SimData structure in memory. Avoid write/read
cycles with simulation data.

Fixed a bug in specfn.py that wrote the spec files multiple times
unnecessarily!

Flake8 changes to hnn_qt5.py

Instead of reading files the old way with np.loadtxt(), use read_* functions
from hnn_core to load data into objects. Reduce the number of file reads by
storing objects in memory until they are used for plotting.

DEP: no longer write spk.txt and rawspec.txt

Update visdpole.py and visspec.py to displays windows within the main
Qt application instead of starting a separate process.

Fixes pep8 errors in modified files (except hnn_qt5.py and paramrw.py)

TODO items:
 - Read dipole and spec data from memory instead of from disk
 - Revisit the case when canvas is initialized multiple times.
   Make sure there isn't a memory leak.

Pass sim_data to the viewer dialog windows instead of reading
dpls from disk again.

Also prepare visrast.py for running in the main Qt app.

Also cleanup code for pep8

MRG hnn_core integration improvements in code

…smainak-fix_import

MRG: Flake8