Experiment with single precision

[maximumcats]

We should compile StarLord in single precision and compare results to double precision (looking at both performance and relative accuracy). Main difficulties are:

• We sometimes use large numbers in the code like 1.d200, which could be replaced with HUGE()
• In astro we usually represent physics with numbers that are larger than the single precision floating point range, for example white dwarf systems where the total energy is O(1e50). A solution here is to use scaled numbers in the actual simulation (say, scaled by the maximum value on the domain in the initial conditions) and undo the scaling only for I/O when we need the absolute numbers.