Fix off-by-one error in iteration counter

[johnomotani]

Bug introduced in #2534. After #2534, calling the monitors always increments the Solver::iteration counter. When monitors are called before the first internal timestep (if dump_on_restart=true), we do not actually want to increment the counter - this call is either at the very beginning of a simulation, when the counter should remain 0 until the end of the first output step, or it is effectively a repeat of the last call to monitors in the simulation that is being restarted from so that incrementing the counter would be double-counting.

Simplest fix seems to be to decrement the counter if the monitors are called at the beginning of a run. Was not an issue before #2534 because the counter was incremented by each solver implementation rather than by the monitors, so the initial call to monitors did not touch the counter.