Enhance marshaling performances

[macisamuele]

This PR mimics what has been done in #336 for the unmarshaling process. As for the unmarshaling also in case of the marshaling process we have a good performance improvements.

Raw data: https://gist.github.com/macisamuele/ee3b66176becab678aedf943e7a0e512

• the worst performance improvement is ~61% (if in the branch takes 1 then on master takes 2.58)
• the best performance improvement is ~72% (if in the branch takes 1 then on master takes 3.60)
• the average performance improvement is ~79% (if the branch takes 1 then on master takes 4.89)

[coveralls]

Coverage decreased (-0.4%) to 97.382% when pulling 8e30cda9e194610ad6e28954df949179e661f108 on macisamuele:maci-enhance-marshaling-performances into 665ca96a5cef81265132ebd0fad09112b0a2f081 on Yelp:master.

[macisamuele]

@sjaensch I've updated the gist content and the description. Performance improvements (according to the tests/profiling/marshal_profiler_test.py) are bound between 61% and 79%