Allow using a Colvars input state file instead of binary checkpoint in GROMACS

[giacomofiorin]

See discussion at: https://gitlab.com/gromacs/gromacs/-/merge_requests/3671#note_1577988430

This PR implements that requirement, but in a different manner than was mentioned before: a new keyword internal to the Colvars config file that specifies a state file to be used unless a state is read in any other way, as provided by the MD engine. After the a state is read the first time, the default input state file name is deleted (so that it won't be loaded again).

The new keyword is currently not tested in CI yet. I feel that it would be confusing to expose it to engines other than GROMACS, which don't need it. Currently, it's only documented for that engine. @jhenin What are your thoughts on TinkerHP?

Note below how the two engines that provide their own restart file differ significantly:

• GROMACS needs a way to continue a simulation when a Colvars config file is changed or when a different build of GROMACS is used (both of which prevent restarting Colvars from the checkpoint file). The proposed change would be to use a default state file for Colvarskeyword colvars-inputstatefile in the MDP, which will be superseded when a checkpoint file is read.
• LAMMPS currently allows (since Unformatted (binary) checkpointing #558) restarting from a Colvars input state file or the LAMMPS restart file. When both are provided, the former overrides the latter, i.e. this behavior is opposite to what seems most appropriate for GROMACS.

While I finish the implementation of this feature for GROMACS, I'd appreciate feedback on whether the above inconsistency should be resolved or not. Mainstream LAMMPS does not contain yet the Colvars code currently in master, so there is time to change.
EDIT: see here #610 (comment)

[giacomofiorin]

Note: converting to draft because the first commit undoes the TPR -> EDR change, this is a conflict that needs to be resolved.

[jhenin]

The issue is, LAMMPS lets the user specify new options for each run even after a restart, so if they mention a state file, it should have priority status - whereas Gromacs doesn't, so if a cpt file can be read with success, the state file could be obsolete from a previous run.

[giacomofiorin]

Merge conflicts resolved, so I was able to add a needed change, i.e. preventing the GROMACS proxy from setting the initial step of the simulation when it doesn't know it anyway. I think that should be addressed separately, but out of scope for this PR.

Also removed the LAMMPS tag since we agreed that the apparent inconsistency is not confusing.

[jhenin]

Great! Just two suggestions on the documentation.

[giacomofiorin]

@jhenin I made the doc edits and restricted the new keyword only to the GROMACS >= 2024 case.

I leave open in case you have suggestions about TinkerHP, otherwise I'm good with merging.

[jhenin]

Colvars/Tinker-HP reads a colvars.state file based on its prefix. I don't think there are specific needs on that side.