Support open-ended ranges in getInputStream, fail loudly on unreadable zip archives

[normanrz]

Fixes #100 and #101.

HttpStore.getInputStream and open-ended ranges (#100)

getInputStream(keys, start, end) formatted bytes=%d-%d unconditionally. Since Store.getInputStream(keys) delegates to getInputStream(keys, 0, -1), that became Range: bytes=0--1 — which no server answers, so the method returned null. It now emits bytes=<start>- when end is negative, mirroring what get(keys, start) already did.

S3Store.getInputStream built the same range string and had the identical bug, so it is fixed the same way (and now validates a non-negative start, matching the other stores).

ReadOnlyZipStore silently reporting an unreadable archive as empty (#101)

ensureCache() treated a null stream from the underlying store as a successfully-parsed empty archive, so a read failure surfaced as valid-but-empty data — concretely, a misleading No Zarr group found at … several layers from the cause. It now throws StoreException.readFailed(...), as it already did for an IOException while parsing the ZIP directory. A genuinely empty archive still parses to an empty index through the normal path, so the two cases stay distinguishable.

Tests

• StoreTest.testInputStreamOpenEnded — shared test asserting the default getInputStream() returns the full object; runs against every store subclass.
• HttpStoreTest.testOpenEndedRangeHeader / testBoundedRangeHeader — MockWebServer tests pinning the exact Range header (bytes=0- and bytes=1-3).
• ReadOnlyZipStoreTest.testUnreadableArchiveThrows — asserts StoreException for a missing archive.

Verified against the reproducer from #100 (https://static.webknossos.org/misc/6001240.ozx):

getSize: 39374104
get(k,0): 39374104
get(k,-22): 22
getInputStream(k,0,-1): ok, first bytes read=4
zip children:  zarr.json 0 1 2
MISSING ARCHIVE throws: Failed to read from store '.../does-not-exist.ozx' at key '': Could not open the ZIP archive from the underlying store

🤖 Generated with Claude Code