Fix regression in a query involving an aggregate function on a

non-wide decimal column in the HAVING clause.

In buildAggregateColumn(), if an aggregate function (such as avg)
is applied on a non-wide decimal column, we were setting the precision
of the resulting column as -1. This later down in the execution got
converted to 255 as in some cases, precision is stored as uint8_t.
The predicate operations on a DECIMAL column has logic that uses
the wide Decimal::s128value field if precision > 18. This logic incorrectly
used the Decimal::s128value instead of the correct value stored in the
narrow Decimal::value field, since precision of the Decimal column
was 255. The fix is to set the aggregate column precision to
datatypes::INT64MAXPRECISION (18) in buildAggregateColumn() when the
aggregate is applied on a non-wide decimal column.

This commit also partially fixes -Wstrict-aliasing GCC warnings.

utils/common/hasher.h

@@ -353,7 +353,7 @@ class Hash128
 public:
     inline size_t operator()(const int128_t i) const
     {
-        return *reinterpret_cast<const uint64_t*>(&i);
+        return (uint64_t)((int64_t) i);
     }
 };