it seems to be some kind of leftover GC-related detritus

it seems to be some kind of GC-related mess

Closes https://github.com/rust-lang/rust/issues/18343

Closes https://github.com/rust-lang/rust/issues/18314

Fixes `config.mk` so that it should not contain multiple inconsistent entries for the same option.

Used aforementioned variants to extract options that have explicit `putvar` calls associated with them in the subsequent code.  When the explicit `putvar` call was conditional on some potentially complex condition, moved the `putvar` call out to the main control flow of the script so that it always runs if necessary.

----

As a driveby fix, captured the error exit when doing the test run of `rustc --version` from `CFG_LOCAL_RUST_ROOT`, and signal explicit configure failure when it did not run successfully.  (If we cannot run `rustc`, we really shouldn't try to keep going.)

----

Fix #17887.

Used aforementioned variants to extract options that have explicit
`putvar` calls associated with them in the subsequent code.  When the
explicit `putvar` call was conditional on some potentially complex
condition, moved the `putvar` call out to the main control flow of the
script so that it always runs if necessary.

----

As a driveby fix, captured the error exit when doing the test run of
`rustc --version` from `CFG_LOCAL_RUST_ROOT`, and signal explicit
configure failure when it did not run successfully.  (If we cannot run
`rustc`, we really shouldn't try to keep going.)

----

Finally, in response to review feedback, went through and identified
cases where we had been calling `putvar` manually (and thus my naive
translation used `opt_nosave`/`valopt_nosave`), and then verified
whether a manual `putvar` was necessary (i.e., was each variable in
question manually computed somewhere in the `configure` script).
In cases that did not meet this criteria, I revised the code to use
the `opt`/`valopt` directly and removed the corresponding `putvar`,
cleaning things up a teeny bit.

----

Fix #17887.

Let's try if not running LLDB tests in parallel solves the sporadic deadlocks we've seen since enabling the LLDB test suite. Running the tests in parallel has lead to unstable behaviour in the past (with LLDB versions below 310.x.x). Maybe our new minimum LLDB version isn't quite up to it either.

cc @alexcrichton