Disable frequency chunking for full-Jones solves

This MR fixes disables the frequency chunking used during the full-Jones solves, as these solves use the MODEL_DATA column which cannot be used with frequency chunking. Such chunking is not needed in any case for the full-Jones solves, as it is a direction-independent solve and thus the memory usage is very low.

For future reference, the error raised by DP3 when frequency chunking is used in this case is the following:

std exception detected: The column MODEL_DATA has shape NCORR: 4, NCHAN: 480, NBL: 1891, while the input buffer for the current step has shape NCORR: 4, NCHAN: 82, NBL: 1891.
Any operation which alters the number of channels, correlations can cause a shape mismatch (example: filter, average, ...). Also operations which will change the metadata can cause a mismatch (example: phase shift, upsample, station adder, ..)

rapthor/lib/observation.py

@@ -326,9 +326,20 @@ class Observation(object):
                     freqchunk_filename = self.ms_filename
 
             # Divide up the bandwidth if needed
-            samplesperchunk = int(round(target_chunksize / self.channelwidth))
-            freqchunksize = samplesperchunk * self.channelwidth  # Hz
-            nfreqchunks = max(1, int(np.ceil(self.numchannels * self.channelwidth / freqchunksize)))
+            if solve_type == 'fulljones':
+                # Don't use frequency chunking for full-Jones solves, as it is not
+                # supported when solving is done against the MODEL_DATA column. The
+                # chunking is disabled by setting samplesperchunk = 0, as this sets the
+                # value of nchan in the DP3 call (and nchan = 0 means all channels; see
+                # https://dp3.readthedocs.io/en/latest/steps/Input.html#msin-46-nchan).
+                # Such chunking is not needed in any case for the full-Jones solves, as it
+                # is a direction-independent solve and thus the memory usage is low
+                samplesperchunk = 0
+                nfreqchunks = 1
+            else:
+                samplesperchunk = int(round(target_chunksize / self.channelwidth))
+                freqchunksize = samplesperchunk * self.channelwidth  # Hz
+                nfreqchunks = max(1, int(np.ceil(self.numchannels * self.channelwidth / freqchunksize)))
             setattr(self, f'nfreqchunks_{solve_type}', nfreqchunks)
             self.log.debug('Using {0} frequency chunk{1} for the {2} gain '
                            'calibration (if done)'.format(nfreqchunks, "s" if nfreqchunks > 1 else "", solve_type))