Tracking Issue for `download-rustc` improvements

[onur-ozkan]

These changes are necessary to make download-rustc more stable and less risky:

• Based on this idea from @RalfJung, instead of checking specific directories to see if they have changed, check for any changes outside of certain allowed directories. This is more maintainable over time because the current approach requires constantly updating hard-coded denied paths.
• When not running in CI, if the "library" tree has changed but the "compiler" tree hasn't, avoid recompiling the "compiler" tree. This approach is similar to how download-rustc=true handles library changes. The goal is to make if-unchanged useful for library developers as well.
• Once above is done, make download-rustc=if-unchanged default option for the "library" profile.
• download-rustc has no effect when running run-make tests #131947

[RalfJung]

When not running in CI, if the "library" tree has changed but the "compiler" tree hasn't, avoid recompiling the "compiler" tree.

Isn't this premature? We can do this once the compiler uses the beta library, because then the library folder truly does not affect the stage 0 compiler artifacts. But today, library is an input to the stage 0 compiler artifacts. Also today, library developers can do ./x.py test --stage 0 library/.. so they can already avoid building the compiler.

[onur-ozkan]

Yeah I forgot mentioning #119899.