Fix dask compatibility in blackbody functions

[djhoese]

I was writing the MERSI-2 L1B reader and needed to use the rad2temp function. I noticed that when dask arrays were passed they were being converted to a numpy array. This PR fixes the usage with the least amount of changes possible.

The biggest difference is that in the planck function a masked array was being used with -9 as the original sentinel value. As far as I could tell the masked array was being used for two reasons: to make min/max calls ignore the masked values and to later use the .mask to replace masked values with NaN. I switched this to use NaN right away instead of -9 and to only print out min/max debug information when the input is an numpy array.

I'm marking this as a bug fix because although its a feature, it is more about fixing a working function so it returns what I want.

• Closes Fix blackbody code to work with dask arrays #73
• Tests added
• Tests passed
• Passes git diff origin/master **/*py | flake8 --diff
• Fully documented

[coveralls]

Coverage increased (+0.3%) to 73.619% when pulling deb43c5 on djhoese:bugfix-blackbody-dask into fd296c0 on pytroll:master.

[adybbroe]

LGTM

[adybbroe]

@pnuu Can you have a look as well before I merge?

[pnuu]

Sorry for the delay. Looks good to me.