Floating point loses precision

[dcleblanc]

An AI review of the code found an issue where some large floating point numbers will lose precision.

For example, a number that uses 54 bits will lose a bit when converted to double because a double only has 53 bits for the value. I feel like this is by design, you asked for a floating point. You get into the same sort of problem when converting from double to say uint64_t. The converted value will leave some number of zeros as the least significant bits.

Proposed fix:

Add (yet another) compile time flag that governs whether it will be strict about floating point conversions, and if the flag is set then turn these into exceptions.