Improve handling of exponentiations

[tohrnii]

Codegen for something like a^b where both a and b are constants results in something like:

(E::from(a_u64)).exp(E::PositiveInteger::from(b_u64))

Ideally it should be something like:

E::from(a_u64.pow(b))

and eventually this should be further optimized to just be

E::from(8_u64)

where a = 2 and b = 3.

This will involve making changes to both IR and codegen where evaluating the exponent before generating code could be an optimization pass.

Originally proposed by @bobbinth in #136 (comment)

[jan-ferdinand]

Not that I have any knowledge of the current status nor any say in this, but constant folding sounds like a good idea in general, not just for exponentiation.

[tohrnii]

@jan-ferdinand Thanks for pointing it out. Agreed, constant folding is something we plan to do as part of adding some optimization passes to the IR in the future.

[bobbinth]

Superseded by #333.