[with patch, with positive review] more NTL wrapping, coefficient access and factoring

[jbmohler]

The attached patch provides some more lines in ntl/decl.pxi to give direct access to some more of NTL. The are two main functionality improvements:

1) __getitem__ now has one less copy of the coefficient. This results in a noticable speed improvement for coefficients (and __hash__)

2) New _factor_ntl and _factor_pari in sage/rings/polynomial/polynomial_integer_dense_ntl.pyx enable us to have more fun with head-to-head comparisons of their factoring routines. It really seems a mixed bag about who is faster. I have a basic check in place that keeps the factoring all in NTL for small coefficients, but this really isn't the end of the story as far as the benchmarking goes.

# original
sage: R.<x>=ZZ[]
sage: f=x^2-1
sage: timeit f.factor()
1000 loops, best of 3: 784 µs per loop

# patched
sage: R.<x>=ZZ[]
sage: f=x^2-1
sage: timeit f.factor()
10000 loops, best of 3: 153 µs per loop

[jbmohler]

a patch with better decision process for pari or ntl

[mhansen]

The patch applies, builds, and passes tests for me.

[rlm]

merged in 2.9.1 rc2