Add Symplectic Integrator

[johannes-spies]

Add a fixed-point iteration based integrator.

[frostedoyster]

The examples/ folder is probably a good idea, but it's too messy at the moment and it needs to be cleaned up. I see some dataset processing code and an ipynb that shouldn't be there.

The symplectic route should also be made available in ASE

[frostedoyster]

There is no Bp or Ap in the sequence

[johannes-spies]

I have to look at the theory for that one. Nevertheless, the functional wrap_* code is just a generalization of what was available in ipi.py all along and the comment just got copied from here

flashmd/src/flashmd/ipi.py

Lines 249 to 250 in 44676ec

	# The barostat here needs a simpler splitting than for BZP, something as
	# OAbBbBABbAbPO where Bp and Ap are the cell momentum and volume steps

Or did I miss something when copying?

[johannes-spies]

Thank you very much for your feedback! I tried addressing all your suggestions. Because the PR became very big, here is a short overview of some key design changes.

• I simplified the example such that it's a single examples/al/al.py file. It's now written in a similar "interactive" Python style, similar to cookbook recipes. tests/test_example_al.py runs a stripped down version (in terms of epochs, simulation steps, and model size) to verify that the example is working, even in CI.
• To support different AtomisticSteppers, there is a new directory available in src/flashmd/steppers. This is also where the original FlashMDStepper moved to. In addition, there is also a SymplecticStepper. This also has the effect, that some parts (for instance, the thermostat versions in src/flashmd/ase or the wrappers) now have to take an AtomisticStepper and not just a model.
• To allow different steppers in src/flashmd/ipi.py, I changed the API a bit and introduced the "wrappers" in src/flashmd/wrappers. They contain the functional core bits that lived in the various get_(nve|nvt|npt)_stepper functions before. These functions only allowed changing the model, which was unsuitable for having different kinds of steppers. The code in the wrappers is essentially copy-paste of what the original get_(nve|nvt|npt)_stepper functions did, but a bit more functional.
• I moved the shared conversion routines to src/flashmd/utils.py. Now, I hope it's easier to convert between various representations of systems: metatomic, ASE, i-PI, and raw position and momenta vectors.

I hope this helps with reviewing this PR!

[johannes-spies]

closed in favor of #43