Introducing multiple precision differential algebra

[darioizzo]

Using the mppp class real128 we here add one gdual to the available ones as to be able to compute with 32 digits of accuracy.

[darioizzo]

I have now to halt to work on this PR and wait for a few upstream changes.

• [x] mppp real 128: https://github.com/bluescarni/mppp/pull/140
• [x] to solve the issue on error: no matching function for call to ‘piranha::polynomial<mppp::real128, piranha::monomial<short unsigned int> >::subs(const string&, mppp::real128) const’ (maybe in piranha?)
• [x] decide what to do with serialization of mppp::real128
• [x] #30 To be merged (still a CI issue there)

I left the code in a compiles and work status commenting out the troublesome lines, i.e. exposition of the gdual in python, and erf for mppp::real128