hdf5 writer: revise file structure

This MR revises the HDF5 file structure and the way that data is stored in the file.

For arrays of variable size (which is not known beforehand), creating individual data sets is easier and faster then resizing operations (tested on array sizes expected on RedPitaya, i.e. 1..16k uint16/32 arrays). The data type can be sent along the meta information as dtype key, otherwise, uint8 will be assumed. Using numpy's view method, these can be recast to arbitrary types.

The payload for start transition is made more explicit, addressing #67

Closes #83 (closed) (obsolete), affects #71, #30