In order to determine which blobs are new based on a given revision, we
call Gitaly's `ListNewBlobs` RPC. This RPC is rather inflexible though:
it only allows passing a single new revision. As a result, we cannot
batch computation of new blobs across a set changes in our access
checks.

As a first step towards fixing this limitation, we now migrate
`#new_blobs()` to use the `ListBlobs` RPC: this is a much more flexible
variant of `ListNewBlobs`, and most importantly it allows us to pass a
set of revisions. When the old implementation is removed, we can thus
easily allow `#new_blobs()` to receive multiple revisions.

Note that this changes the return type of `#new_blobs()`: instead of
returning a set of `Gitaly::NewBlobObject` classes, we now return a set
of `Gitlab::Git::Blobs`. While both share the same `size` and `path`
attributes, the former tracks the blob ID via an `id` attribute while
the latter uses an `oid` attribute. There is only a single callsite of
`#new_blobs()` though, which is the FileSizeCheck push rule, and this
callsite only uses the `size` and `path` attributes. So in the end, this
change in behaviour is fine.