Rewrite fmt! for performance

[brson]

fmt was written in the stone ages and is not fast. It should be rewritten to use either an io::Writer or some kind of string builder, and it should use stack allocation everywhere it can. It should not emit lots of inline code to bloat up function bodies, instead delegating to core::extfmt functions marked with #[inline], leaving the amount of inlining up to LLVM.

[brson]

It should also probably use traits to work with more types, per #1653

[catamorphism]

Not critical for 0.6, de-milestoning

[catamorphism]

This doesn't seem to have been nominated, but had a milestone attached. Does this really need to be on production-ready? Is fmt! usually a huge performance bottleneck? Re-nominating so we can discuss.

[graydon]

accepted for backwards-compatible milestone

[alexcrichton]

This is currently underway with the new ifmt! syntax extension. While not completely done, I think that the remaining progress of that extension is better tracked in the #2249 metabug.