[lohedges]

Currently loch only samples water, and the osmotic (μpT) ensemble isn't valid for water as the solvent of a dilute system, such as a solvated protein, which is loch's main use. Water's chemical potential is then almost that of pure water, so the number of waters is barely fixed, and any mismatch between the excess chemical potential and the water model's chemical potential drives the total water count steadily one way. Sampling only within the GCMC sphere doesn't avoid this, since water diffuses between the sphere and the bulk.

The osmotic ensemble is valid for water only when its chemical potential depends strongly on how much is present, e.g. adsorption into a porous solid or water in a non-aqueous solvent. These aren't loch's target use.

Changes:

• Remove the pressure argument and the per-move box and Adams value updates. A barostat is rejected again, as before osmotic support was added.
• _compute_adams_values() is called once at construction, using the box from set_box(). The long-range correction reuses that box volume. set_box() no longer accepts an OpenMM State.
• Remove tests/test_osmotic.py, and restore the README note that GCMC must be performed at constant volume, removing the μpT documentation (including Fix issue #59 #60).