st_curvetoline maxError instead of segments/quarter paramter

[mjurce]

Let the function compute the number of segments per quarter respect to the error approximation given from the user. The number of segments should dpend on the circle curvature. The fixed parmeter "segm/quarter circle" is not always good with big and flat arcs.

The error parameter could be the max distance between the segment and the arc.

[strk]

Implementing this would allow to support conversion of 3-points lines to curves #2627

[mjurce]

The maxError " maxDist(chord,arc) with n points segmentation could be computed: maxError=r*(1-cos( theta/(2*(n+1)) ) )

Solving for "maxError" should give:

nPoints = theta/(2*acos(1- maxError/r))

Someone could check if this is correct? Then is easy to plug-in into arc segmentizer. The angle increment will be given from this formula.

[strk]

For the record, this is implemented in QGIS: ​https://qgis.org/api/classQgsCircularString.html#ab11654b8e0e12ad86f98a0d5dd70adee

[strk]

Work started in ​https://gitlab.com/postgis/postgis/merge_requests/3

[strk]

In 15430:

Implement extended ST_CurveToLine signature

Adds lwcurve_linearize function at liblwgeom level.
Turns lwgeom_stroke into a wrapper, keept it for backward compatibility.
Reduces allocations in linearization procedures.

Implements SYMMETRIC and RETAIN_ANGLE flags.
Implements MAX_DEVIATION, MAX_ANGLE and SEGS_PER_QUADRANT configs.

Includes unit and SQL tests.
Includes documentation.

Closes #2464 (maxError configuration is MAX_DEVIATION)
Closes #3772 (balanced output is SYMMETRIC and RETAIN_ANGLE flags)

Document the new ST_CurveToLine signature