Updated pari to 2.7.6.

Done for version 2.7.6 (released 20/06/2016):
[last column crossreferences current development release 2.8.0]

  Fixed
    1- [from 2.7.5] O(1) == O(x) returned 0                      [#1765]
    2- matsolve(a,b) and a^(-1) gave wrong results [or SEGV] when t_MAT  [F86]
      a was not square and a,b "modular" (F2m,Flm,FpM,FqM,F2xqM,FlxqM)
      same for x^(-1)                                          [#1666]
    3- e=ellinit("11a1"); ellztopoint(e,3*e.omega[1]/5)[2] wrong [#1683] [F97]
    4- K.codiff incorrect if [K:Q] > 2                                  [F130]
    5- nfrootsof1(polcyclo(85)) -> 85 instead of 170            [#1766] [F133]
    6- at \p19, polroots((x+1)^2*(x-1)^7*(x^2-x+1)^5*1.0)-> SEGV [#1767] [F134]
BA  7- ellsea returned the trace instead of the cardinal as documented.  [F135]
BA  8- ellsea(,,1)  could return a wrong result                  [#1768] [F136]
MC  9- memory leak in pari_close: sopath not freed                      [F138]
  10- isprimepower(30011^(3*17)) returned 0                            [F141]
BA 11- [mpi] interrupt/alarm could cause a crash                        [F143]
BA 12- [mpi] relinking empty t_LIST caused a crash                      [F144]
BA 13- my(s=1,a=0);forstep(i=1,20,s,s++;a+=i);a -> wrong result          [F146]
  14- sqrt(0) -> loss of accuracy (sqrtn was correct)                  [F148]
  15- K = bnrinit(bnfinit(y^2-5),[1,[1,1]]); bnrdisc(K) ->wrong [#1804] [F152]
  16- ellztopoint(ellinit([-1,0]), I) -> wrong result          [#1800] [F153]

(wiz)