Fix several issues preventing reading state from LAMMPS restart file

[giacomofiorin]

With both NAMD and LAMMPS now allowing to first initialize Colvars after a simulation has begun (initial step > 0), it's getting harder to get everything right for moving restraints that require knowledge of the step when they were first defined (see #486).

This PR fixes this behavior by recording that number as firstStep and embedding it in the state file.

For LAMMPS specifically, this now works both with a Colvars state file provided via the input keyword or by the LAMMPS binary restart file.

[jhenin]

Looks good. Why is the new function delete_input_stream necessary? What happens with close_input_stream?

[giacomofiorin]

Looks good. Why is the new function delete_input_stream necessary? What happens with close_input_stream?

It just rewinds the string stream to allow future reads (as when the files are read as strings embedded in a TPR file), but then the condition below remains true, and the input state string keeps being loaded every time setup_input() is called:

colvars/src/colvarmodule.cpp

Lines 1313 to 1320 in 0a997fc

	if (proxy->input_stream_exists("input state string")) {
	cvm::log(cvm::line_marker);
	cvm::log("Loading state from string.\n");
	read_restart(proxy->input_stream("input state string"));
	cvm::log(cvm::line_marker);
	proxy->close_input_stream("input state string");
	}

The above is clearly an issue that needs to be fixed, either by adding a new function (as in this case) or a separate flag that is activated once and deactivated after the state is read. This is where a second set of eyes is most helpful ;-)