Built-in floating point alloc functions

[ranweiler]

Right now, theft provides theft_random_double(). This returns a double in the closed unit interval [0, 1].

What would be nice is some other floating point alloc builtins that do things like:

• Take a user-defined range of output values
• Produce only "bad" values with different floating-point classifications
• Produce only arbitrary finite values
• Produces literally any valid IEEE 754 floating-point value, at the relevant precision

[silentbicycle]

In particular, the naming seems important here -- theft_random_ieee754_double, for example.

theft_random_double should probably be deprecated/renamed in a future release.