Implement mode in backends

[jonasteuwen]

This implements a mode flag in the SlideImage backends. This is required since the ICC profiles need to know what the output type is (e.g. RGB or RGBA). While OpenSlide always returns RGBA, this is not the case for libvips.

[BPdeRooij]

The mode flag implementation seems like a good solution. An error will be automatically raised if either inMode or outMode is not supported when building the ImageCmsTransform. ICC were correctly applied when tested on a TCGA image for both PyVipsSlide and OpenSlideSlide backend.

I do have some comments/questions about the PyVipsSlide backend:

1. The mode can already be set when initializing PyVipsSlide. For _read_as_openslide this can always be "RGBA". For _read_as_tiff a combination of self._images[0].get("interpretation") (documentation) and number of bands can be used.
2. Slightly off-topic, but related: could we use pyvips.Image attributes self._images[0].get("xoffset") and self._images[0].get("yoffset") as offset in the slide_bounds for _read_as_tiff?
3. libvips reserves "icc-profile-data" for the ICC profile. ICC profiles are also stored in this property when pyvips.Image.tiffload is used. After some testing, I believe the NotImplementedError is not needed when using "tiffload".

[jonasteuwen]

I don't know about self._images[0].get("xoffset"), we should test that. We might also need to write those, so if we can write mrxs -> tiff we could use something like this if it works.

I don't think it will always be RGBA with openslideload, this will depend on the image you're loading (in contrast with openslide-python)

[BPdeRooij]

I believe images will be loaded as RGBA when using openslideload, but checking the number of bands in images will also generate the correct results in any case. The changes look good to me.