add dtype argument to combine-function.

[MSeifert04]

Fixes #391

This PR adds a dtype parameter which is passed to Combiner.

This fixes two Bugs:

• if the dtype of the first ccd passed to combine isn't float64 the result was downcast to the dtype of the first ccd.
• in that case also the memory consumption calculation was wrong because Combiner used float64 while the calculation assumed the dtype to be the same as the one of the first ccd.

And also allows to optionally reduce the memory consumption by simple using a dtype.