This had to be broken out of #7375 because it turns out python_path is used on windows to communicate the location of the Python interpreter even when it's not otherwise available on the PATH at execution time.

We can possibly switch this bit of client-server communication to use a hidden flag in place of --python_path. Alternatively (and probably preferably), we can replace this logic with a repo rule that detects the host platform's Python interpreter, and feed that into the Python toolchain. Another option is to deprecate this bit of niceness and instead require that the interpreter either be available on PATH at execution time or else have its path embedded into an appropriate py_runtime of a toolchain definition.