Reduce precision to eliminate duplicate points from InflatePaths

[charlieb]

Fixes #91607 by locally reducing the precision of InflatePaths to the Clipper2 default. This prevents duplicate points from being generated and thus it doesn't cause decompose_polygon_in_convex to fail.
This also replaces pull request 91627

[rburing]

I'd like to restore this to the value it was before my PR

• Replace Clipper1 library by Clipper2 library #90153

which accidentally increased the precision.

Before that PR it was 0.25f * (real_t)SCALE_FACTOR where SCALE_FACTOR = 100000.0, and now we have the constant PRECISION = 5 which is related to SCALE_FACTOR by SCALE_FACTOR = pow(10, PRECISION), so we should now use 0.879588001734407 * PRECISION (or rounded 0.88 * PRECISION) to restore the previous value, since $$(\log(10^5 \cdot 0.25)/\log(10)) / 5 \approx 0.879588001734407.$$

Please test if that also solves the problem.

Edit: Actually 0.25 * SCALE_FACTOR was used as the arc_tolerance.

[charlieb]

@rburing
The InflatePaths interface specifies an int for the precision parameter so 4.8 isn't going to work unless I pull the InflatePaths function apart like I did in this branch.

I'm wondering if it's possible to just change the #define PRECISION to 4 instead of 5 and call it done? Do the other functions really need 0.00001 which would be equivalent to 10 microns?