RFC: Infer the type of float literals

[brson]

Roadmap says "There is general consensus that floating point literals are too long; 1.0f will likely become sugar for 1.0f32. This is purely a backwards-compatible change."

Not sure the details - I think we already have a f literal though for float.

[pcwalton]

The fact that f already means float is a problem. I propose keeping the literals the way they are and inferring which float is meant, much as we do for integers.

[catamorphism]

Done in fe02814