Readers give incorrect error message for unknown calibration

[simonrp84]

Describe the bug
If a user selects an incorrect calibration type for the abi_l1b reader then the error message they receive is incorrect.

To Reproduce

from satpy import Scene
from glob import glob

files = glob('*.nc')
scn = Scene(files, reader='abi_l1b')
scn.load(['C04'], calibration='counts')

Expected behavior
An error message is produced that says: Unknown calibration counts.
This is defined here.

Actual results
The following error message is given:

KeyError                                  Traceback (most recent call last)
<ipython-input-12-34ce17a0197c> in <module>
      1 print(scn.available_dataset_names())
----> 2 scn.load(['C04'], calibration='counts')

/network/aopp/apres/users/proud/satpyconda/lib/python3.7/site-packages/satpy/scene.py in load(self, wishlist, calibration, resolution, polarization, level, generate, unload, **kwargs)
    997         if unknown:
    998             unknown_str = ", ".join(map(str, unknown))
--> 999             raise KeyError("Unknown datasets: {}".format(unknown_str))
   1000 
   1001         self.read(**kwargs)

KeyError: 'Unknown datasets: C04'

Environment Info:

• OS: Ubuntu 18.04
• Satpy Version: 0.21.0

Additional context
Running with an accepted calibration type (such as calibration='radiance') works successfully.

[djhoese]

This is likely not specific to ABI. I think the Unknown datasets message is only using what you've provided for the names ['C04']. So the message isn't necessarily wrong since counts is a calibration that Satpy knows about but C04 doesn't have that calibration: the C04 dataset, as requested, is unknown.

We could make the error message longer and include any of the other parameters that were provided (calibration, polarization, etc). That may have to change depending on how @mraspaud's dynamic DatasetIDs work.

[simonrp84]

Yes, I suspected it might be a bit more broad than just ABI. Would be nice to have it fixed, though, so that the user gets a more accurate error message.
Or, on the other hand, delete the code in the ABI reader and improve coverage ;)