default null ordering is opposite of PostgreSQL

[xiangjinwu]

Describe the bug
We do not support specifying nulls {first | last} yet. When nothing is specified, the default behavior of PostgreSQL is to treat nulls as larger than non-nulls (i.e. nulls last for asc and nulls first for desc).

But in our current implementation, memcomparable encodes nulls as smaller than non-nulls. And parts not using memcomparable are implemented to keep consistent with it rather than PostgreSQL.

To Reproduce

create table t (v1 int);
insert into t values (2), (null), (3), (1);

select * from t order by v1;
select * from t order by v1 desc;
select * from t order by v1 limit 2;
select * from t order by v1 desc limit 2;
create materialized view mv_a as select * from t order by v1;
create materialized view mv_d as select * from t order by v1 desc;
create materialized view mv_a2 as select * from t order by v1 limit 2;
create materialized view mv_d2 as select * from t order by v1 desc limit 2;
select * from mv_a;
select * from mv_d;
select * from mv_a2;
select * from mv_d2;

Expected behavior
Same behavior as PostgreSQL.

Additional context
Do not see an immediate need to fix it (before fully supporting nulls {first | last}). Just documenting the current behavior. And we should fix the default behavior when we implement nulls {first | last}.