Consider optimising the`stitch_neurons_mst` function by computing the mst only on the fragmented nodes like below: https://github.com/connectome-neuprint/neuprint-python/blob/b4dfd061bb12bf8d4834b9db05727b44cbd29c59/neuprint/utils.py#L469