Floating-point redesign

[KFAFSP]

Current floating point is fixated on IEEE-754, which is already not enough for future usage scenarios.

Design considerations:

• s/q NaN distinction should not be modeled
• NaN and Inf may remain special values, but must follow our own spec everywhere (no minNumber confusions)
• Floating-point types should be product types
• Floating-point environment functions that are independent of rounding (ldexp, frexp, ...) should be the minimal set of required interpretation functions
• We should consider a binary_float ancestor type

[KFAFSP]

• Define NaN, Inf semantics outside of IEEE-754
• Remove special handling of NaN, Inf where now inappropriate