[mypyc] Generate smaller code for casts

[JukkaL]

Merge a cast op followed by a branch that does an error check and adds a
traceback entry. Since casts are very common, this reduces the size of
the generated code a fair amount.

Old code generated for a cast:

    if (likely(PyUnicode_Check(cpy_r_x)))
        cpy_r_r0 = cpy_r_x;
    else {
        CPy_TypeError("str", cpy_r_x);
        cpy_r_r0 = NULL;
    }
    if (unlikely(cpy_r_r0 == NULL)) {
        CPy_AddTraceback("t/t.py", "foo", 2, CPyStatic_globals);
        goto CPyL2;
    }

New code:

    if (likely(PyUnicode_Check(cpy_r_x)))
        cpy_r_r0 = cpy_r_x;
    else {
        CPy_TypeErrorTraceback("t/t.py", "foo", 2, CPyStatic_globals, "str", cpy_r_x);
        goto CPyL2;
    }

The generated assembly is the same in the successful case, but assembly for the
error handler is more compact, reducing the size of binaries.

We could also do a similar thing to unbox ops in the future.

[msullivan]

Great cleanup