Update bignum to CPAN version 0.65

Update bignum to CPAN version 0.65

    [DELTA]

 * Restore the upgrading and downgrading in "bignum", with the exception
   of upgrading from Math::BigFloat to Math::BigRat, which I couldn't make
   sense of. Now, both integer literals and integers (and Infs and NaNs)
   that are the result of a computation become Math::BigInt objects, and
   both non-integer literals and non-integers that are as a result of a
   computation become Math::BigFloat objects. The classes used for literal
   integers and non-integers during compile-time are configurable, as well
   as the classes used for downgrading and upgrading during runtime.

 * Add the "bigfloat" pragma, which converts all numeric literals to
   Math::BigFloat objects. This is consistent with the "bigint" pragma,
   which converts all numeric literals to Math::BigInt objects, and the
   "bigrat" pragma which converts all numeric literals to Math::BigRat
   objects.