Use of np.float32 over np.float64 in calibration factors might undesirably affect the values of SEVIRI brightness temperature

[pkhalaj]

This has been raised by @simonrp84, and has been recently resurfaced as a result of numpy 1.x to numpy 2.x transition. See for example #2983.

@simonrp84 states (verbatim) that

Regarding the work to ensure satpy uses np.float32 instead of np.float64: I get a bit worried about how this will work for the brightness temperatures. For example, with SEVIRI is we are forcing the radiances and all the other calibration factors into 32-bit then that could have undesirable effects when we compute the brightness temperature (as we have a lot of very small and very big numbers in the equation). Has anyone done a sensitivity study on that to ensure the BTs aren't affected?