BatchedFetchStore for shard index caching and range batching

[wietzesuijker]

Wraps all fetch stores with BatchedFetchStore to cut HTTP requests when rendering sharded Zarr v3 datasets.

zarrita.js creates a fresh shard-index cache per open(array) call, so for a 3-band RGB layer the same shard index is fetched 3× per shard per viewport. BatchedFetchStore caches suffix-length range reads (always shard-index fetches) in an LRU so bands 2 and 3 are served from memory. It also batches concurrent getRange calls within the same event-loop tick: calls are accumulated, sorted by offset, grouped into merged ranges (32 KB gap threshold), fetched as fewer HTTP requests, then sliced back to callers — modelled on geotiff.js BlockedSource.

For a 3-band sharded Zarr v3 (e.g. Sentinel-2 EOPF, 9-tile viewport):

	Before	After
Shard index requests	3× per shard per viewport	1× (cache hit for bands 2+3)
Chunk range requests	N sequential GETs per band	≤ ⌈N/group⌉ merged GETs per path

BatchedFetchStore is adapted from the zarrita.js development branch (MIT, same org) and should be removed from this repo once @zarrita/storage ships it.

AI (Claude) supported my development of this PR.

[vercel]

@wietzesuijker is attempting to deploy a commit to the carbonplan Team on Vercel.

A member of the Team first needs to authorize it.