NStarOffsets testing & refactor

[GermanHydrogen]

Tasks:

• [x] Tests
• [x] Refactoring

Fixed:

• Wrong default value if Gaussian fit fails https://github.com/pyobs/pyobs-core/blob/d5f79525842a7ff8cad0e650dd10b867385865de/pyobs/images/processors/offsets/nstar.py#L359

Problems:

• Box-Size is capped at 5x5 pixels which seems to small -> Change max_offset to pixel https://github.com/pyobs/pyobs-core/blob/9d0feb85ca8a6ccbd788a30c73731df9a2f447ab/pyobs/images/processors/offsets/nstar.py#L80
• Default Box-Size is incompatible with photutils.psf.extract_stars, which expects odd box-sizes https://github.com/pyobs/pyobs-core/blob/9d0feb85ca8a6ccbd788a30c73731df9a2f447ab/pyobs/images/processors/offsets/nstar.py#L111
• The mean of the box offsets seems to be a bad choice for a location parameter as it is not robust. The median or $\alpha$-trimmed mean would be a better choice. -> Different cases for n https://github.com/pyobs/pyobs-core/blob/9d0feb85ca8a6ccbd788a30c73731df9a2f447ab/pyobs/images/processors/offsets/nstar.py#L297
• Unconventional background contrast calculation. -> Extract to filter https://github.com/pyobs/pyobs-core/blob/9d0feb85ca8a6ccbd788a30c73731df9a2f447ab/pyobs/images/processors/offsets/nstar.py#L218
• median squared relative residue (?) as a metric for fit quality https://github.com/pyobs/pyobs-core/blob/9d0feb85ca8a6ccbd788a30c73731df9a2f447ab/pyobs/images/processors/offsets/nstar.py#L366C2-L366C48

[GermanHydrogen]

• Changed the unit of the class parameter max_offset to pixel
• Seperated the source filter from the box generator and created a new processor ImageSourceFilter
• Changed the background contrast metric to weber contrast

After further thought, I did not change the location parameter for calculating the offset from the box offsets. I also did not change the fit quality metric. Both things should be first tested under real conditions, to determine if they are sufficient.