Reimplement NUMERIC datatype using base-10000 arithmetic; also improve

some of the algorithms for higher functions.  I see about a factor of ten
speedup on the 'numeric' regression test, but it's unlikely that that test
is representative of real-world applications.
initdb forced due to change of on-disk representation for NUMERIC.

src/test/regress/expected/aggregates.out

@@ -2,15 +2,15 @@
 -- AGGREGATES
 --
 SELECT avg(four) AS avg_1 FROM onek;
-        avg_1        
----------------------
- 1.50000000000000000
+       avg_1        
+--------------------
+ 1.5000000000000000
 (1 row)
 
 SELECT avg(a) AS avg_32 FROM aggtest WHERE a < 100;
-       avg_32       
---------------------
- 32.666666666666667
+       avg_32        
+---------------------
+ 32.6666666666666667
 (1 row)
 
 -- In 7.1, avg(float4) is computed using float8 arithmetic.
@@ -120,9 +120,9 @@ group by ten order by ten;
 (10 rows)
 
 SELECT newavg(four) AS avg_1 FROM onek;
-        avg_1        
----------------------
- 1.50000000000000000
+       avg_1        
+--------------------
+ 1.5000000000000000
 (1 row)
 
 SELECT newsum(four) AS sum_1500 FROM onek;

src/test/regress/expected/numeric.out

@@ -665,9 +665,9 @@ SELECT t1.id1, t1.result, t2.expected
 -- ******************************
 -- numeric AVG used to fail on some platforms
 SELECT AVG(val) FROM num_data;
-         avg          
-----------------------
- -13430913.5922423207
+          avg           
+------------------------
+ -13430913.592242320700
 (1 row)
 
 -- Check for appropriate rounding and overflow