Switched the code to a float from uint64 so that precision could remain strong
even when the values are tiny. This means we can apply decay even when the
values are tiny.

Test updated to reflect the clamps and newer decay methods and strengths.