Default runners have 2 CPUs. We can achieve a nice speed-up
(but not quite linear) by leveraging runners with more CPUs.

Way back when we only supported statically linked extension modules. And
we had to disable `_testcapi` because it wouldn't build statically.

Now that we support building extension modules as shared libraries,
it makes sense for extension modules in support of tests to not be
part of libpython and to be shared libraries.

This commit makes that change.

And since `_testcapi` is buildable as a shared library, we enable it.
In doing so, we eliminate the largest single source of test failures.