reproject_and_coadd bug with circular data

Hi,
I'm working on the Hi-GAL dataset which cover the whole Milky Way and I want to generate a global fits map from the 37 available mosaics.
In order to do that I use the *reproject_and_coadd* method:
```
merged_data, _ = reproject.mosaicking.reproject_and_coadd(hdu_list, new_header, reproject_function=reproject.reproject_interp)
```
But it is not fully working:
- One to two mosaics aren't merged with the other mosaics.
- The mosaic(s) which aren't merged is (are) always the one(s) the furthest from the fits reference pixel.

I think the issue come from the circularity of the data: since every mosaic overlaps on the left and right mosaics, this produce a maps wider than 360° (in galactic wcs). I would guess the *reproject_and_coadd* method is working properly until the merged map is close to 360°, let's say we merged mosaics until it covers from 5° to 355°. Then when adding the last mosaic covering from 354° to 6°, I think the *reproject_and_coadd* method consider (based on galactic coordinates) than the new mosaic is covering 6° to 354°, hence is superposed with the merged one instead of adding the new mosaic either to the left or to the right.
I hope this is clear enough to make my problem understandable.
Thanks for any kind of help,
Loris Berthelot

Start with the reproject.mosaicking.reproject_and_coadd call shown in the report and reproduce it using the 37 Hi-GAL mosaics and the supplied galactic WCS. Investigate how inputs near the 0°/360° boundary are mapped and determine why the outer mosaic is omitted; done means all mosaics contribute to the global FITS map.

Written by the indexing model from the issue text.