Python wrappers are not profiled separately when binding=True

[mhsmith]

Cython 0.27.3, Python 3.6.3.

#cython: binding=True, profile=True

cpdef f(x):
    if x > 0:
        f(x - 1)

>>> cProfile.run("for i in range(5): proftest.f(2)")
         23 function calls (8 primitive calls) in 0.000 seconds

   Ordered by: standard name

       ncalls  tottime  percall  cumtime  percall filename:lineno(function)
            1    0.000    0.000    0.000    0.000 <string>:1(<module>)
         20/5    0.000    0.000    0.000    0.000 proftest.pyx:3(f)
            1    0.000    0.000    0.000    0.000 {built-in method builtins.exec}
            1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}

By comparison, with binding=False (the default):

>>> cProfile.run("for i in range(5): proftest.f(2)")
         28 function calls (18 primitive calls) in 0.000 seconds

   Ordered by: standard name

       ncalls  tottime  percall  cumtime  percall filename:lineno(function)
            1    0.000    0.000    0.000    0.000 <string>:1(<module>)
            5    0.000    0.000    0.000    0.000 proftest.pyx:3(f (wrapper))
         15/5    0.000    0.000    0.000    0.000 proftest.pyx:3(f)
            1    0.000    0.000    0.000    0.000 {built-in method builtins.exec}
            5    0.000    0.000    0.000    0.000 {built-in method proftest.f}
            1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}

[mhsmith]

Because the statistics of the function and its wrapper are merged, the result is that every cpdef function which is called from Python will be reported as having recursive calls, whether it's actually recursive or not.

[scoder]

Probably easy to fix. My guess is that the DefNodeWrapper generates different code than the normal DefNode (in Nodes.py), or something like that. Probably also worth reconsidering if tracing the "call" and "return" events in both the wrapper and the function itself is a good idea.
PR welcome.

[scoder]

As noted in #3611 (comment), when using type specs for extension types, the wrapper is apparently being profiled, leading to failures in pstat_profile_test.pyx (where it's being counted twice).

Probably related or something to fix together.

Worth testing if #3138 resolves this, too.