cf. issue #8

Extended the code to incorporate two-particle bonds.

clog() only logs on the root rank.

Probably add ArgumentParser for --verbose to control logging output.

1. echos the conf key value pairs
2. add --destdir, --profile, --verbose
3. reformat output files to standard 5 digit beam shard names:
   https://beam.apache.org/releases/javadoc/2.2.0/org/apache/beam/sdk/io/ShardNameTemplate.html
   (assuming we will run to 1M ranks)

This is buggy but at least allows me to run the code.

Logging exchange cost and use new pmesh exchange routines.

Use the standard python package directory layout. Add utils/ directory
for auxiliary scripts. Add a placeholder docs/ directory for future
sphinx autodoc deployment.

Adds a requirements.txt file to manage python package dependencies.

Also update the README with how to build for different use cases.

Also update build instructions in README

Ensure that github-linguist does not incorrectly classify any OCCAM
input/output files as ROFF text-formatting files

Revert commit 29a56a4940d71db8ca25450d868ed769ccef88c4. Turns out it is
necessary that test/ is a python module in order to test without
installing the main source as a package with `pip install -e .` first.
Need to add setup.py, clean up main.py, and make a proper package
before this file can be removed.

This way we avoid opening/closing the file an extra time when reading
the data in main.py.

Add particle bonds, reorganize directory structure, add some testing

If no 'RESPA' or 'respa' keyword is specified in the CONF input,
nothing is changed and the rRESAP scheme reduces automatically to a
normal Velocity Verlet integrator.

If no molecular forces are present (no 'molecules' is found in the
input .hdf5), the rRESPA integrator reduces automatically to a
(inefficient) Velocity Verlet integrator using only the field forces.

Adds the rRESPA integrator of Tuckerman et al.

To make it more convenient to import in other modules.

Until new mpsort, pmesh, pfft-python tags are pushed to PyPI solving
the build issue causing mpi4py, numpy, and cython to be present before
install starts, we can handle the dependencies with a pyproject.toml
file, cf. PEP 518.

Start printing testing code coverage report in build CI

Allow Logger.rank0.log() and Logger.all_ranks.log() for logging output
from root or all ranks, respectively.

While main.py is not pip installable yet, make sure tests can import
from the package root properly.

Ensure we don't keep junk around in the CONF dictionary after we parse
it from CONF.py.

Ensure energies match theoretical Gaussian core model equivalent

Rename hPF/ to hymd/ in anticipation of the PyPI package name, finalize
the movement of functions away from main.py into modules in hymd/. Add
new modules field, file_io, thermostat.

Ensure the warning prints species alphabethized  when chi is specified
for some interaction but one (or both) of the species are not present
in the system.

Allows data to be directly read into VMD using the VMD-h5mdplugin.

This *probably* breaks the --disable-mpio option in main, so only way
to run is with driver=mpio.

Fixes a bug causing domain decomposition to be run every single step,
and adds command line options to selectively disable forces (in the
most stupid way possible, but it does not matter).

Adds a simple makefile to compile the fortran code into python modules.

Prints energies per particle instead of the default total energies.

Hopefully eliminates the need to ever create new Field objects when
calling update_field. Pre-allocates the necessary Field and ComplexField
objects in main before the MD loop starts.

Using double precision for the internal logic in compute_angle_forces to
avoid numerical instability in the calculation of the sine of theta from
the cosine value. Also fix a bug causing the forces to not update
propely.

When rij and rkj are very close to parallel, the denominator in the
force expression diverges as 1/sin(theta). The numerator vanishes
sufficiently rapidly to cancel the divergence, but the numerical
expression is unstable. In the cases where cos(theta) is Nπ to
numerical precision, we just ignore the force and the energy. Mirrors
what is done in GROMACS.