Add error handling to restart routines

[jakelangham]

Restarting simulations works great, unless the output files are corrupted for some reason. This can happen if the simulation was interrupted while writing out, for example.

It will be a pain to do, but the relevant routines in Restart.f90 could use better error handling to treat these situations better. For example, on attempting to restart from output N and failing, the nice thing to do would be to fall back to output N-1 if it exists and can be used. Instead the typical behaviour is to exit or segfault.

[markwoodhouse]

Restarting segfaults if the restarted simulation finished (i.e. reached "T end" or maybe "N out" corresponds to number of output files).

A check should be added -- probably based on "Last output file in RunInfo.txt".

I think this can be included when dealing with the corrupted file issue.