Numerical Data Types

[tensionhead]

Current State and Problem

Right now, we are implicitly (down-)casting everything to double float precision (32bit for real, 64bit for complex). As touched upon in #264, we should at least inform the user about this, and it might be better to also allow the users to keep their original (64bit most likely) precision. From our top of the head discussions, in 99% of the cases this will just lead to double RAM and disc space usage without any significant effect on the scientific results. But there might be cases for certain algorithms ('catastrophic anihiliation') where this really can make a huge difference.

Proposed solution

We need (across all submodules) an interface for importing/working on/saving data which allows for the free choice of the float precision. However, we still want to encourage 32bit precision as the de-facto standard to save resources.