Split _bt_insertonpg to two functions

Here's a patch that:

Moves the logic to find a page with enough room from _bt_insertonpg to a
new function, _bt_findinsertloc. It makes the code more readable, and
simplifies the forthcoming Grouped Index Tuples patch.

Also, the insert location within page used to be calculated twice for
unique indexes, once in _bt_checkunique and second time in
_bt_insertonpg. That's a waste of cycles, and this patch fixes that.

I couldn't measure a difference with pgbench, but this micro-benchmark
shows it:

> psql postgres -c "CREATE TABLE inserttest (i int PRIMARY KEY);"
> psql postgres -c "TRUNCATE inserttest; checkpoint;"; sync
> time ~/pgsql.cvshead/bin/psql postgres -c "TRUNCATE inserttest;
INSERT INTO inserttest SELECT a FROM generate_series(1,1000000) a;"

Without patch: real 0m7.260s
With patch: real 0m6.963s

On my laptop, fsync=off, full_page_writes=off, checkpoint_segments = 10,
to remove any other variables.

It's not a huge difference, but it's worth having, and performance
wasn't the main motivation of the patch anyway.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com