ast/interpreter: Use match statements

[dcbaker]

The slightly increases the lines of code (adding the match statement itself), but generally makes the code easier to follow (and potentially faster), since we can replace things like if struct.attribute == x: with case x: and isinstance(struct, Foo) with case Foo().

I have not converted every if/elif tree into match statements, but have tried to be a bit conservative of only converting the big ones. This also doesn't make use of any super powerful match features, like unpacking.

[bonzini]

Bytecode-wise I see

293           COPY                     1
              LOAD_GLOBAL             10 (str)
              LOAD_CONST               0 (())
              MATCH_CLASS              0
              COPY                     1
              POP_JUMP_IF_NONE        53 (to L4)
              NOT_TAKEN
              UNPACK_SEQUENCE          0
294           POP_TOP

instead of

292           LOAD_GLOBAL              7 (isinstance + NULL)
              LOAD_FAST_BORROW         1 (obj)
              LOAD_GLOBAL             10 (str)
              CALL                     2
              TO_BOOL
              POP_JUMP_IF_FALSE       52 (to L4)
              NOT_TAKEN

where MATCH_CLASS is (see _PyEval_MatchClass in cpython

    assert(PyTuple_CheckExact(kwargs));
    // First, an isinstance check:
    if (PyObject_IsInstance(subject, type) <= 0) {
        return NULL;
    }
    // Short circuit if there aren't any arguments:
    Py_ssize_t nkwargs = PyTuple_GET_SIZE(kwargs);
    Py_ssize_t nattrs = nargs + nkwargs;
    if (!nattrs) {
        return PyTuple_New(0);
    }

So there is no call (though in the end you pay anyway the overhead of ABCMeta.__instancecheck__) but there are a few more bytecodes to be executed.