Hello! While testing AORRTC with non-Euclidean cost spaces, I ran into a couple of leaks and robustness issues:

• Memory leak in AOXRRTConnect::reset(): trees were cleared without freeing the Motion objects they hold by raw pointer. Now calls freeMemory() first. Also frees cmotion->state in solve() cleanup.
• findNeighbour() padding: absolute pad (rootDist += 1e-5) becomes negligible for large rootDist (e.g. SE2, SE3, non-Euclidean metrics), causing empty nearestR results. Replaced with relative pad rootDist *= (1 + 1e-6).

Minor fixes bundled in:

• bounds check in findNeighbour() cost-iteration loop
• initialise GrowState gs = TRAPPED
• fix tgi.xmotion bookkeeping in solve()'s straight-line check
• inner connect loop now respects ptc