Remove much of the crate file evaluation infrastructure

[brson]

Having a mini scripting language inside the parser is rather undesirable, and some of it's use case will be replaced with the new conditional compilation attributes. Try to remove much of the front::eval code for parsing crate files.

Graydon suggests that leaving const definitions and variable substitutions is 'probably harmless'.

[graydon]

Other guidelines welcome. I find it difficult to draw a line here. I don't want to have a full shell language in here, but it feels like crates are a "natural" place to do some configuration-management logic. Maybe just conditionalizing by #[cfg()] will be enough. It's almost certain that any real build will have a CM system outside the crate as well, invoking rustc. But huge and non-obvious compiler-argument lists are something from the C world I hope to avoid.

Maybe if the interface between outside CMs and inside rustc is always just "a list of cfg variables" and you can pass that in a file, that's adequate?

Perhaps give rustc a --cfg-file= option. And/or the ability to override with -D foo=bar for cases where you have something against using config files?

[brson]

I've removed everything that wasn't being used, and that includes the ability to set and evaluate variables. I figure that if the need arises again it is easy to restore.

[brson]

Graydon, I've continued the discussion of how to merge configuration files into the compilation environment in #612.