[folbricht]

An index server answers a request for an index that doesn't exist with 400 Bad Request unless its store is a local directory. A client reading from such a server gets unexpected status code 400 instead of the missing-object error it gets from a server with a local store.

The handler used os.IsNotExist to tell a missing index from a failure to read one. That only matches the error a local store returns. The other index stores report a missing index in ways it doesn't recognise:

Store	Error for a missing index
HTTP, OCI	NoSuchObject
SFTP	the not-exist error flattened into a new one with %v
GCS	storage.ErrObjectNotExist, wrapped
S3	none from GetIndexReader; the S3 error surfaces from the first read

Changes:

• NoSuchObject matches fs.ErrNotExist through an Is method, so errors.Is(err, fs.ErrNotExist) covers a missing index in any store, wrapped or not.
• The index handler uses that test for GET and HEAD.
• The SFTP and GCS index stores return NoSuchObject for a missing index.
• The S3 index store reads ahead in GetIndexReader, which is when the object is actually requested, and returns NoSuchObject for NoSuchKey. This takes no extra request. It also corrects HEAD on an index server backed by S3, which answered 200 for any name since opening the reader never failed.

Other failures to read an index are answered with 400 as before.

The handler, the S3 store and the SFTP store have tests for a missing index. The GCS change has none, there is no fake GCS backend in the tests.