Fix HDF5Header for Python2/Python3

There is a problem which bubbled up recently, when using aanet with Python 3. The raw_header table is built from strings and in Python 3 those are unicode. However, pytables converts the dtypes of the Table to numpy.bytes_ and writes an HDF5 dataset with that dtype. In the next iteration of the pipeline, the data is then tried to be appended to the dataset and the dtypes mismatch (unicode vs bytes).

This needs to be harmonised for both Python 2 and 3, which is a bit awkward... I'd rather ditch Python 2 forever to avoid this annoying work but I guess we have to live with that for a few months...