Improve raster selectivity estimator

[strk]

As discussed in #1828 the selectivity estimator does noo support functional comparison, but for the raster case functional is the only possible index, and the histogram is already based on that exact same function.

This ticket is to find out if it's possible to check if the function used in the query is the same used to build the index. When that's the case, the estimator should behave exactly as if the query was not functional.

[strk]

I've moved first steps toward support for functional index estimation: ​https://github.com/strk/postgis/tree/svn-trunk-raster-estimator Just in case you want to join [ not sure I'll ever continue it ]

[strk]

Continued and completed ! r12814 in trunk

As a bonus track, we now support selectivity estimation on every expressional index whose expression returns a geometry.

[robe]

refer to #2851

I think debbie I can work around by recompiling her postgres with xml support. Can we be sure most people testing will have their dev postgres compiled with xml support? I need to do the same for winnie (9.2 and below since I test those thru mingw).

the winnie regress not sure what causes that might be an issue.

[strk]

See #2895 for a "can warm" consequence of this