ST_GeneratePoints performance regression

[pramsey]

By my testing, ST_GeneratePoints seems to have gotten about 20% slower in 3.5 versus 3.4. There was some churn in the code to fix a failure, but things that were meant to make it faster seem to have done the opposite.

WITH circle AS (
        SELECT 'POLYGON((1 0, 0 1, 15 16, 16 15, 1 0))'::geometry AS geom
),
pts AS (
        SELECT (ST_Dump(ST_GeneratePoints(geom, 50000))).geom AS geom
        FROM circle
)
SELECT 'generate_points_1', count(pts.geom)
FROM circle, pts;

[pramsey]

Turns out this was a phantom generated by the PARANOIA setting making building collections very very Slow.