Reduce the Rust stack red zone

[brson]

Right now the stack red zone is 20K on all platforms, which is ~20K larger than we would like. Several considerations:

• The dynamic linker performs lazy binding in the red zone. Lazy binding can be turned off.
• Landing pads call _Unwind_Resume in the red zone. We can probably come up with a scheme to switch stacks for this, but even better will be to implement our own unwinding.
• There is some logging happening in the red zone that only shows up with RUST_LOG set appropritiately.
• There is at least one upcall still doing a lot of work in the red zone.
• We need to be able to handle signals in the red zone, haven't put any thought into it yet.

[pcwalton]

For the last one, what about sigaltstack(2)?

[jckarter]

For some platforms you could handle the last point with sigaltstack, but (afaik) there isn't an equivalent to sigaltstack on Windows, and sigaltstack + pthreads is busted on some lesser platforms, OpenBSD at least.

[msullivan]

brson says that the linux red zone, at least, is down to 2k.

I'm marking this as milestone 5.

[thestinger]

This may or may not still be an issue with a new segmented stack implementation, so it can be closed for now.