Remove modes from libstd and libcore at least, enable deprecated mode warning by default

[graydon]

[graydon]

Some other components of this bug:

• Fix main to not take args, just have a sys::get_argv or something
• Fix the Eq and Cmp traits
• Add a -Z flag for compiling with the old mode logic; turn old mode logic off by default, only on for rustc (and/or old libraries if they don't get done in time)

[bblum]

why not have main's optional arguments be &[~str] instead of &&~[~str]?

[jdm]

One benefit of a sys::get_argv is that it reduces the "public static void main(String[] args)" factor.

[brson]

I prefer to keep the main argv signature, mostly because sometimes it's useful to call main manually, and not having an argv in the signature means involving global runtime state. For example, the jit is currently tested by building crates and calling main.

[nikomatsakis]

I agree with @brson. We were discussing this the other day and I remembered how much I dislike how Python does it. I always end up making a main(args) fn anyhow. Basically it's bad for all the normal reasons global state is bad—it's unclear what the inputs to the fn are, and it gives the caller no control over them.

[catamorphism]

Current status: serialization and serialization2 cause several mode warnings, and net_tcp causes one warning. Two others (in ptr and extfmt) will go away once I make a new snapshot. This is otherwise done.

[catamorphism]

Still warnings from (in core) stackwalk::frame_address, and (in std) net_tcp and serialization. Otherwise done. Deprecated modes forbidden in all modules except these, AFAIK.

[catamorphism]

I'm closing this because we've done all we're going to do for 0.4; the ones I mentioned in my previous comment have been muted.