Added single precision floating point f32

[Viriathus1]

Added single precision float as type f32 which was mentioned in #305.

Slightly different implementation from the other numerical primitives in that all operations call the C backend directly through WASM exported functions as discussed.

[antocuni]

@Viriathus1 thank you a lot for this.
It seems mostly complete, but it lacks two things IMHO:

1. automatic coercion for ieee754_div, see my comment below
2. conversion rules for f32.

For example, the following two calls should work, but they fail:

def foo(x: f32) -> f32:
    return x

foo(1)
foo(1.2)

TypeError: mismatched types
  | /tmp/x.spy:7
  |     foo(1)
  |         ^ expected `f32`, got `i32`

TypeError: mismatched types
  | /tmp/x.spy:7
  |     foo(1.2)
  |         |_| expected `f32`, got `f64`

similarly, I should be able to convert from f32 to int if I do it explicitly:

def foo(x: f32) -> int:
    return int(x)

TypeError: cannot call objects of type `type`
  | /tmp/x.spy:4
  |     return int(x)
  |            |_| this is `type`

[Viriathus1]

@antocuni I've added the changes for this one. Sorry I was a bit lazy with the coercion rules mostly because it would be resolved in this issue #333. Let me know if I've missed something

[antocuni]

good choice here! I think that saturating conversion is what happens in rust and C#, so +1 to do the same in SPY

[antocuni]

@antocuni I've added the changes for this one.

thank you, it looks very good now, merging!

Sorry I was a bit lazy with the coercion rules mostly because it would be resolved in this issue #333. Let me know if I've missed something

no problem :).
I fear it will take a while to fix #333 because it needs some serious language design to understand how to model the "type of a numeric literal before it's coerced to a concrete type".