Implement support for dylib linkage

[daivinhtran]

This PR implements dylib linkage against the standard library behind a feature flag --@rules_rust//rust/settings:experimental_use_dylib_linkage.

The main part of this feature is here where we skip exporting static rust stdlibs and export only libstd.so instead.

This feature is useful when the subset of libstd being statically linked to downstream shared libraries and binaries is larger than the entire dylib version of libstd. The following diagram is the high level of what dylib linkage is trying to achieve.

Running the feature against android_binary yields a size reduction on the shared library produced by android_binary because it doesn't statically link the rust stdlibs anymore.

> bazel build //:android_app --config=android_x86_64 
> unzip -l bazel-bin/android_app.apk
Archive:  bazel-bin/android_app.apk
  Length      Date    Time    Name
---------  ---------- -----   ----
  1381968  2010-01-01 00:00   lib/x86_64/libandroid_app.so <--- static link with rust stdlibs
---------                     -------
  1390294                     9 files

> bazel build //:android_app --config=android_x86_64 --config=dylib_linkage
> unzip -l bazel-bin/android_app.apk
Archive:  bazel-bin/android_app.apk
  Length      Date    Time    Name
---------  ---------- -----   ----
     8080  2010-01-01 00:00   lib/x86_64/libandroid_app.so <--- reduced size because of dynamic linking
 13055776  2010-01-01 00:00   lib/x86_64/libstd-8d416d49cf02ecea.so
---------                     -------
 13072400                     10 files

Here, the benefit comes when there are enough shared libraries statically linking against the rust stdlibs. "Enough" here means that the total up size of those libraries being more than just the entire libstd.so.

TODO: I'm leaving this PR without unit tests until I get some feedback or suggestions on my approach.

[daivinhtran]

@illicitonion I added analysis tests and fixed the flag name. PTAL.

The tests fail on window and darwin for not having prefer-dynamic set here. For some reason, it only happens on window and darwin. The test overrides the flag explicitly.

Does anyone know why?

[illicitonion]

I'm afraid nothing jumps out about why the tests are failing, but happy to spend a bit of time debugging if no one else has any insights!