It's been confusing that these tests have been called "interop" but
are not actually tests of interoperation.

This reverts commit 3045a379.

Backports per-object grpc_init/deinit and separated-file grpc protoc
codegen (#7538, #8246, #8920).

The server cleanup method is untested.

The join() function that exposes it is only called by the internals of threading.py, and we don't hold a reference to the server thread to explicitly join() it, and I'm not sure we should add a reference just for this purpose.

Moreover, the threading.py only calls join(None), the code path in question isn't even exercised by the internals of threading.py. Its just there to make sure we properly follow the join(timeout) semantics.

To support magical internal build processes, the pb2 files need to be
split into pure-proto and gRPC parts. This performs that split and
further fixes bad module names in the test harness that interfered with
the intended test implementation.

An unfortunate side effect, due to limitations of protoc and holdover
behavior we must support in major version 1.x, is that trash files are
generated in split generation. This shouldn't be a problem in normal
protoc plugin use.

Some features relating to proto2 extension querying are missing due to
missing upstream features required to support them (see
https://github.com/google/protobuf/issues/2248).

Moves all tests into a separate package. This does not change existing
supported means of running tests (e.g. through run_tests.py).

When the Python Interpreter exits, it first attempts to join any
outstanding threads.  This is problematic if a server is created
as a top-level variable or referenced by a reference cycle, as join()
will hang.  This adds cleanup behavior to the server thread's join().