flaky: OMPL freespace examples under xdist — re-seed race aborts planning (~1/700)

[jf---]

known-class flake, recording it now that the windows full suite (landed via #105) runs the OMPL examples under xdist and inherits it:

• symptom: test_freespace_ompl_example / test_lowlevel_freespace_ompl_example fail with Planning failed: ErrorTask: Error (Abort Triggered); pass in isolation and on rerun
• telltale stderr: Random number generation already started. Changing seed now will not lead to deterministic sampling (ompl RandomNumbers.cpp:80)
• mechanism: an xdist worker that already ran an OMPL test re-seeds mid-process; sampling goes non-deterministic and the planner occasionally misses its time budget. direct descendant of the flaky: OMPL-pipeline example tests fail stochastically (unseeded RNG -> TOTG negative path velocity) #103 work (per-test RNG_setSeed, single-planner first-solution config) and the dropped determinism contract test — seeding alone can't beat the ParallelPlan race
• observed: ~1 in ~700 test executions (once locally on macos arm64, once on windows CI test-wheel 3.12, same day; siblings green both times)

candidate fixes, none urgent: bump the example's planning time budget (cheapest), serialize ompl-example tests onto one worker (xdist_group), or revisit upstream determinism. parking with evidence so the next person who sees Abort Triggered lands here instead of re-deriving it.