Fix facet correction when not applying beam

There was a separate code path for the case that I/XX/YY was imaged, and primary beam was disabled, and facet solutions are enabled. In that case, it can be avoided to write the primary beam images to disk because they can be applied in situ. However, this code doesn't work with Q/U/V, and even Stokes I seems to correct the trace instead of diagonal. Doing the correction is complicated so I think it is better to incur the writing penalty and have one code path that can do everything. That's what's done in this MR.

This also fixes an issue that Bjorn is having.