match bootstrapped compiler to architecture

[becker33]

@scottwittenburg noticed that when bootstrapping compilers, we just concretize the compiler spec without trying to match its arch/os to the spec it came from.

This PR ensures that the bootstrapped compiler is built for the same architecture as the spec for which it is being bootstrapped.

[tgamblin]

2 nitpicks @becker33:

1. Should we be this specific, or should we use the family? I'm just curious because I thought there was a discussion where when we detect compilers we add the family as the target.
2. What about cross-compiled environments? Should we use the frontend family if the target is the backend? Given that I'm not sure we can handle that situation properly now anyway I am not sure I care for this PR, but I thought I'd ask.

[becker33]

We could put os=fe target=fe in, although that would require a bit more refactoring. Not too much though.

[becker33]

@tgamblin on point 1, the target of a compiler and the target for which a compiler package is built are different things. We definitely want the compiler target to be the family, because that controls which processors it knows about for cross-compilation. But some users probably want their compilers built for the fastest architecture they can, to speed up build times. I don't know how much of a difference in build time it makes to build the compiler for broadwell vs x86_64, but it could be enough to be worth something.

[tgamblin]

I think I will merge this for now (cc: @scottwittenburg), and this may not even be an issue when compilers are dependencies. I definitely think there is some stuff we need to flesh out for compiler modeling.

The way we currently handle the target attribute is not quite as clear about where the compiler runs and what it targets as I'd like it to be -- I think we use the target for both...