More informative XYZ file reading errors

[jhenin]

especially in VMD:

• give details about discrepancies in number of atoms found and expected
• quit early after an error to avoid drowning error messages in later, irrelevant output
• close files upon errors so they can be reopened once errors are corrected (critical in VMD Dashboard)

In principle, could be extended to similar situations with file IO, but the benefits are less obvious than this use case.