Finish the polish on format!

[catamorphism]

As per FIXMEs in syntax::ext::fmt, refactor nested functions inside pieces_to_expr, and also move validation code from make_new_conv into something in core::extfmt.

[catamorphism]

Also: decide how important it is to be printf-compatible and whether parse_type should support two different "signed" specifiers.

[pcwalton]

I plan to rewrite fmt as part of the new I/O design.

[msullivan]

I am renaming this bug and promoting it to the metabug for fmt! redesign. I am moving it up to milestone 2, because we may wish to change fmt in non backwards compatible ways.

Linking all of the fmt! issues I have found:

Major rewrite/redesign:

• Rewrite for perf: Rewrite fmt! for performance #3571
• Support internationalization: String formatting and internationalization #4630

Feature requests:

• More flexible arguments: #fmt strings should allow index of argument, count, and precision to be specified #1651
• Support %e: fmt! does not support %e #6593
• Support %g: extend fmt! to add %g analogous to printf '%g' #3810
• Allow different width arguments: #fmt should allow any width int, float, etc to be used #1653
• Optimize for certain arguments: fmt! could be optimized for some combinations of arguments #6035

Bugs/misfeatures:

• Doesn't require commas between args: fmt! does not depend on commas between arguments #4982
• Reject weird combinations of specifiers: #fmt accepts some combinations of precision and 0 padding that gcc warns about #2481
• We use different default precision of floats than printf: #fmt uses wrong default precision for float #1375

[huonw]

Linking to the Lib fmt wiki page.

[alexcrichton]

Changing name from "Rewrite fmt!" because that's mostly been done and this is now polish.

[huonw]

(I updated @msullivan's comment to a task list, and it seems to work.)

[catamorphism]

De-milestoning. Enough of the work is done for now.

[brson]

Obsolete.