Fix read and write from pmb.df

[pm-blanco]

Solves #102 and #114 finally.

Changed

• pd.NA is now enforced as value for empty cells in pmb.df, this prevents transformation of variable types from int to float which was breaking the code when reading pmb.df from file (See Integer values in pmb.df become float when reading a pyMBE dataframe from file #102).
• Espresso bond objects stored in pmb.df now retain the same value for the ._bond_id attribute as the original Espresso objects.

Fixed

• Writing and reading pmb.df from file does no longer change the variable type from int to float when there are empty cells in the column.

I had to do quite a bit of refactoring to make the code work with pd.NA because it behaves differently than np.nan, but I checked that all functional tests still run.

@paobtorres @1234somesh this PR solves the issues you raised!

[jngrad]

To solve the bookkeeping of ._bond_id I had to hack a bit the espresso bond object because otherwise it was hard to backtrack it.

You may want to discuss it with @RudolfWeeber, he is planning to rewrite the bond id management in the next release of ESPResSo. It would be best if we could update pyMBE at the same time we update the bond id management in ESPResSo.

[pm-blanco]

You may want to discuss it with @RudolfWeeber, he is planning to rewrite the bond id management in the next release of ESPResSo. It would be best if we could update pyMBE at the same time we update the bond id management in ESPResSo.

Should we discuss it today during the ESPResSo coding day or better in our meeting scheduled next week?

In principle the issue with ._bond_id is not breaking the functionalities of pyMBE at the moment (at least that we are aware of) so I could revert the patch and wait for the next ESPResSo release to fix it. We could also leave the patch and adapt it to the new framework once the new version of ESPResSo is released, it is actually not too many lines of code that are involved.

[pm-blanco]

@jngrad I tested the handling of the _bond_id private attribute of bond objects in espresso and it seems to be robust. Bonds seem to retain the same _bond_id, even if you delete one bond and afterwards you save and load a checkpoint. See a minimal working example below.

Should then keep this PR as it is, or you think is better that we remove support for tracking of _bond_id in pyMBE for now?

import espressomd
from espressomd import interactions
from espressomd import checkpointing

restart=True
# Set-up the checkpoint
checkpoint = checkpointing.Checkpoint(checkpoint_id=f"mycheckpoint", 
                                      checkpoint_path=".")

if restart:
        checkpoint.load()
    
else:
        Box_L = 7.5*pmb.units.nm
        espresso_system = espressomd.System(box_l = [Box_L.to('reduced_length').magnitude]*3)
        espresso_system.time_step=0.1
        checkpoint.register("espresso_system")
        bond    = interactions.HarmonicBond(k   = 0,
                                                r_0 = 1)

        bond1    = interactions.HarmonicBond(k   = 1,
                                                r_0 = 1)
        bond2    = interactions.HarmonicBond(k   = 2,
                                                r_0 = 1)


        bond_id = espresso_system.bonded_inter.add(bond)
        bond_id1 = espresso_system.bonded_inter.add(bond1)
        bond_id2 = espresso_system.bonded_inter.add(bond2)

        for item in espresso_system.bonded_inter.items():
                print(item)

        espresso_system.bonded_inter.remove(bond_id1)
        checkpoint.save()

for item in espresso_system.bonded_inter.items():
     print(item)

[jngrad]

Besides the requested change, LGTM. The bond id business will need to be revisited when we refactor the bond storage mechanism in ESPResSo, but that project is months away.