Various Rich SIMD Reconstruction improvements

Miscellaneous updates and bug fixes.

• Replace the lookup table for log(exp(x)-1) with a new fast approximate form. This is a nice example of where one should not take the call grind results too serious as according to that tool the lookup table was much better, with less instructions, but it had terrible memory cache issues which caused it to in the real world be much slower than what has replaced it, which is fast calculation (based on an approximate log implementation and a power series expansion.)
• Convert a number of transient TES data objects from double to float.
• Fix some FPEs.
• Reduce the number of iterations in the newton raphson solver (still more than good enough for float).
• Various smaller improvements.

Note, these changes will cause some perturbations in the nightlies tests.