Performances:
- use of the slightly faster pandas.read_csv instead of np.genfromtxt
- only keep a limited number of scalar fields in memory (attribute
  nfields_max of StagyyData). This greatly reduces the memory
  consumption when a large number of snapshots are processed

Bugfixes:
- fix tstart and tend options of time subcommand
- tstart option default value is None instead of 0. to properly handle
  runs with a negative time

Internals:
- misc.InchoateFiles context manager uses fnames as default prefix if
  None (default) is specified as tmp_prefix
- setup.py now contains minimum version numbers of dependencies