Fix to my fix

[Roland Dreier <droland@mail.math.okstate.edu>]

My previous fix for the factorization bug had a problem: the power of p we
have to raise things to when taking their pth root might be too big to fit
in a long.  So I should have made frobinv a GEN:

*** src/basemath/polarit1.c.orig        Thu Jul 23 08:40:23 1998
--- src/basemath/polarit1.c     Thu Oct  8 15:01:24 1998
***************
*** 1595,1598 ****
--- 1595,1599 ----
    GEN ex,y,f2,f3,df1,df2,g,g1,xmod,u,v,pd,q,qq,unfp,unfq;
    GEN *t;
+   GEN frobinv = gpowgs(pp,lgef(a)-4);

    if (typ(a)!=t_POL || typ(f)!=t_POL || gcmp0(a)) err(factmoder);
***************
*** 1616,1620 ****
        j=(lgef(f)-3)/p+3; f2=cgetg(j,t_POL);
        f2[1] = evalsigne(1) | evallgef(j) | evalvarn(vf);
!       for (i=2; i<j; i++) f2[i]=f[p*(i-2)+2];
        f=f2; df1=deriv(f,vf); f3=NULL;
      }
--- 1617,1621 ----
        j=(lgef(f)-3)/p+3; f2=cgetg(j,t_POL);
        f2[1] = evalsigne(1) | evallgef(j) | evalvarn(vf);
!       for (i=2; i<j; i++) f2[i]=(GEN) gpow((GEN) f[p*(i-2)+2],frobinv,0);
        f=f2; df1=deriv(f,vf); f3=NULL;
      }

Roland