In the user manual, the functions ROUND and TRUNCATE are said to produce
the same result type as that of the first argument, however that is
not always true. Especially, some types are different when the
second argument (the decimal precision) is const vs. non-const.

This patch adjusts the resolved type to be in better accordance with the
type of the first argument.
For all integer types, the resolved type is BIGINT.
For all floating point types, the resolved type is DOUBLE.
For all non-numeric types, the resolved type is DOUBLE.
For the type DECIMAL, the resolved typs is also DECIMAL.
The type attributes are also copied from the first argument,
except for the type DECIMAL, when the second argument is a constant value:
- When the desired number of decimal places is less than the
  scale of the argument, the scale of the result is adjusted
  accordingly. The precision of the result is also adjusted.
  In addition, for the function ROUND the precision is extended with
  one position to accomodate for rounding that increases the number of
  significant digits.
  Example: ROUND(99.999, 2). Here, the type of the argument is
  DECIMAL(5, 3). The type of the result is DECIMAL(5,2), since the
  result of this operation is 100.00.
- With negative second argument, the type is adjusted to have scale 0
  and a corresponding precision.

Reviewed by: Guilhem Bichot <guilhem.bichot@oracle.com>
Change-Id: Id5af0be433cdd52fc2e7a240de72cdc838d2722e