rustdoc could use some LD_LIBRARY_PATH handling cleanup

[pnkfelix]

Spawned off of #13732.

Much like our run-make tests, rustdoc has a problem where it sometimes wants to be using the host's LD_LIBRARY_PATH to resolve dynamic library dependencies, and sometimes it wants to be using the target's LD_LIBRARY_PATH. (This largely comes up e.g. when running the testable code blocks embedded within rustdoc documentation blocks.)

This issue is not normally exposed under normal uses of rustdoc because most people run it against a stage2 build or a rust distribution. The problem really only arises when running the documentation against stage1.

For the run-make tests, PR #13753 is taking the approach of feeding in two distinct environment variables (one for the host and one for the target) and relying on tooling within the makefile select the appropriate option. Presumably some similar approach would work for rustdoc.

(But in the short term @pnkfelix is just going to disable the incompatible uses of rustdoc for stage1 and point them at this ticket.)

[pnkfelix]

Discussions with @alexcrichton lead us to think that this is not truly just a stage1 problem. As he puts it: "It is just luck that we can substitute the stage2 host binaries for target binaries."

See discussion here: https://botbot.me/mozilla/rust-internals/msg/14338015/

[pnkfelix]

note in particular: https://botbot.me/mozilla/rust-internals/msg/14339992/