Fixing load_seismogram_rdbm

 - Was in a mixed up state between rdbm and kerner. Especially the
   parameter strain_type had a completely different meaning in rdbm and
   kerner, which lead to wrong seismograms being read in with wrong
   amplitudes, which affected kernel amplitudes.
 - Introduce new function load_strain_point_interp_seismogram, which is
   only used for the case of reading seismograms. It's not nice to have
   a very similar functionality than load_strain_point_interp in a
   separate function, but I do not see a way around. Main differences:
    1. It always works in the straintensor_full mode, since we need the
       full strain tensor to retrieve seismograms for moment sources
    2. It does not use any buffers, which might be initialized with the
       wrong dimension, if working with vp kernels (and therefore
       straintensor_trace mode). Since every seismogram is only loaded
       once, and seismometers are usually not collocated, this should
       not affect performance at all.
  - Removed function load_seismogram completely. Was not used in the
    main code anymore, but in some tests. N.B. Its functionality was not
    tested anywhere.