Enable profiling of Cython functions

As it is not possible to profile Cython functions by default, this adds
a few options to ensure that Cython compiles these functions with
profiling support enabled. This way we should now be able to see how
long they take with cProfile as used in our nightly benchmarks.

nightly-benchmark/run-dgx-benchmark.sh

@@ -21,7 +21,7 @@ which python
 
 echo "Cythonize Distributed"
 pushd "${CONDA_PREFIX}/lib/python3.8/site-packages/"
-cythonize -i \
+cythonize -i --directive="profile=True" \
 	"distributed/protocol/serialize.py" \
 	"distributed/scheduler.py" \