Cache getSRSbySRID

[Algunenano]

Trac: https://trac.osgeo.org/postgis/ticket/4672

For ST_AsX3D remove it altogether since it was never part of the output (so no point on calculating it and passing through functions).

Measurements:

- 683788 Points (1 vertex on avg)
-- EXPLAIN ANALYZE Select ST_AsGML(the_geom, 8) from benchmark_7773a711c8441d4b494a51fd9feebeac7a9b9c734619398620293;
-- Latency avg before: 8286.371 ms
-- Latency avg with cache: 388.738 ms
-- Perf: 21.31x


-- EXPLAIN ANALYZE Select ST_AsX3D(the_geom, 8) from benchmark_7773a711c8441d4b494a51fd9feebeac7a9b9c734619398620293;
-- Latency avg before: 38299.269 ms
-- Latency avg with cache: 408.281 ms
-- Perf: 93.80x

-- EXPLAIN ANALYZE Select ST_AsGeoJSON(the_geom, 8, 2) from benchmark_7773a711c8441d4b494a51fd9feebeac7a9b9c734619398620293;
-- Latency avg before: 7720.442 ms
-- Latency avg with cache: 192.258 ms
-- Perf: 40.15x

[Algunenano]

Since we are here, cache also getSRIDbySRS for a 40x in ST_GeomfromGeoJSON (or the equivalent '{}'::geometry):

-- On top of previous change
EXPLAIN ANALYZE Select ST_AsGeoJSON(the_geom, 8, 2)::geometry from benchmark_7773a711c8441d4b494a51fd9feebeac7a9b9c734619398620293;
-- Latency avg before: 32311.245 ms
-- Latency avg with cache: 738.984 ms
-- Perf: 43.72x

[Komzpa]

lwgeom_free(lwgeom); ?

[Komzpa]

Probably okay since we're just before leaving a function and context is going to die anyway, but I nervously rechecked it's in postgis/ and not lwgeom/.

[pramsey]

Anything except index support functions you get the free "deallocation on exit".

[Algunenano]

Yes, I've been doing this lately in multiple postgis functions (the ones were liblwgeom now returns a lwvarlena_t object). I think it keeps things cleaner and it's good for performance as it's faster to wait for the deallocation of the context.