Speed up new navmesh gen by x1.6 - optimize height and alpha terrain sampling to avoid redundant bounds check / deep sea / wrapping / dispatches, have terrain meshing work inplace on the buffers instead of using pooled vertex and index pooled lists - avoid using rust native list, it's not really pooled it just reallocates the internal native array everytime it's unpooled, use custom one (may be overkill, will check later if native array is good enough), bypass pooling these by using [ThreadStatic] buffers - replace cancellation tokens with lightweight tileCancellation classes just wrrapping volatile bool, we just need the object reference to act as a build version to avoid locks and stale results, tokens are massive if allocated per tile and add complexity, and a lot of features go unused