existing generate_loaders() function.

This is a little out of sync still with the wip_hyperopt branch, and I
have yet to resolve the differences between the two, but this helps me
to understand the current functionality and where improvements and
abstractions can be made.

fixes to the README.

builds of pytorch and friends.

Fix DOS line endings on existing environment/requirements files.

Add a brief docstring in simul_data.py that the code there is just for
testing / demonstration purposes and pointed out by Jean; we might
remove it later (though I believe it might be useful for tutorial
purposes and, with some refactoring, for generating a dataset template).

The conda environment file is specifically for a CPU-only setup; need to add a
separate one for GPU once I get on a machine that actually has a CUDA-capable GPU.

simul_data.py script.

This will likely be rewritten in the future, but having this test makes
for a nice demonstration, and a nice check against future changes (it
provides something of a partial certificate of the current functionality
of the code).

module-level global variable.

This will likely be refactored even further to make it easier to
override specific parameters in different contexts.  But this still
makes it easier to override all defaults, esp. for testing.

be what was intended, but never completed.

In particular the intent seemed to be that the scenario_idx column in
generated scenario params be a simple integer, rather than a name like
"scenario_0000".

of the "seed" parameter from the simulation params.

for it to take a value.

Use action='store_true' instead.

argv, rather than defaulting to argv.

Also return an exit code instead of calling ``sys.exit()`` directly.

This makes testing for these functions easier.  This is just a
preliminary move intended to test setting up a test framework.

Enough to be able to `pip install -e .` for more effective development.

package called dnadna.

As many of the modules are also scripts, their
`if __name__ == '__main__'` blocks are converted to `main()` functions,
which are called from executable scripts now living in the `scripts/`
directory.

This `scripts/` directory will mostly go away in the next stage,
however, as we add a `setup.cfg` and implement `console_scripts`
entrypoints, and possibly sub-commands.

Not all of these scripts are currently actually working with each other;
most are in some state of disrepair to be cleaned up later, but with the
exception of cleaning up some imports they are otherwise left unmodified
by this commit.

Also adds the beginnings of a .gitignore file.

Although these may be worth keeping in a repository, I don't believe the
belong in the main sources for a distributable package based on this
work.  Instead they should go in a personal repository which uses this
one as a dependency, or in a specialized branch, etc.

Merge branch 'regression_classif' of gitlab.inria.fr:ml_genetics/private/dnadna into regression_classif