hdf_to_gro.py is very slow due to heavy file read load

[mortele]

Instead of reading from the HDF5 datasets N times, just reading the entire dataset into memory (buffered if the trajectory is huge) and indexing into it will be 100 times faster.