expose and add selectivity to the 3d/4d index (&&&) bindings

[robe]

How could you not ticket such an important feature :).

[robe]

Paul -- done with this? I know you've exposed it at the very least? Though haven't had a chance to stress test it.

[pramsey]

No, this is still to be done, and a significant piece of work.

[strk]

This one doesn't seem in time for 2.0. Probably shouldn't be blocker either.

[pramsey]

Implementation here has also taken in aspects from #2101, to allow easier testing and verification of results. I'm quite happy with the constant selectivity estimates in my tests so far, and the join selectivity is if nothing else more sensitive to actual data distribution than before. All tests pass, including a couple tests for selectivity itself! (regress_selectivity.sql). r10796 on trunk

[pramsey]

Hearing no issues, I declare this complete.

[strk]

This broke the NULL return from ST_Estimated_Extent, see #877 and #3396