Support for SFL (no shift) projection

In this case, only the projection kind is forced to SFL instead of GLS/radio, but not the reference. Data is not modified. We do not expect the absolute positions of the pixels to be correct, but we can compare if the 3 software consistently return the same positions for their generic SFL implementation at $\delta_0 \ne 0$.

• ds9 and astropy.wcs are consistent: they have the same interpretation of the header,
• G $\backslash$CONVERT in GILDAS version jul23 disagree,
• after reimplementation of the SFL projection, G $\backslash$CONVERT in GILDAS version oct23 is consistent with the 2 other ones.