The HAMT exporter loads the whole shard in order to find a leaf

[achingbrain]

If the dir-hamt-sharded exporter finds a link that points to a subshard, it loads that subshard even if it won't contain the path we're trying to load.

The hashes generated by the hamt algorithm are stable, however, so you don't need to load the entire tree in order to find out where a given node will be, instead you can go level by level in the hamt, recreating it as you go. We use this approach in js-ipfs-mfs to add/remove DAGLinks without loading the entire graph.

This module should use a similar approach, otherwise we don't see any benefit from sharding.