Release v2.10.1

Fixed

• Asking for pme: gpu or update: gpu in mdrun_options stopped the
  build at its first minimization. The same options go to every mdrun,
  and GROMACS refuses both tasks on a GPU without a dynamical integrator.
  For minimization stages they now become auto, and so does bonded, which
  hasn't been checked with a minimizer. Dynamics gets the options unchanged.

• The parameterization summary undercounted and named the wrong place.
  It was printed before the symmetry-, chain- and second-shell-derived
  templates were built, so a run that reused 29 templates reported 11. It
  also always named ~/.htpolynet, even when HTPOLYNET_CACHE pointed
  elsewhere. It now comes after every template exists, reports "N of M
  reused", and names the library and cache actually searched.

Changed

• The "percolation not assessed" message says when it means "cannot
  assess". It read as though the outcome were always the correct answer
  for a linear polymer, with the multi-residue crosslinker case as an aside.
  For example 2 that aside is the actual situation: bis-GMA crosslinks
  through junctions a residue-level count cannot see. Noticed by a sweep of
  the shipped examples.