Fix duplicate objects in "thin+cached" packs from DFS

Fix duplicate objects in "thin+cached" packs from DFS

The DfsReader must offer every representation of an object that
exists on the local repository when PackWriter asks for them. This
is necessary to identify objects in the thin pack part that are also
in the cached pack that will be appended onto the end of the stream.

Without looking at all alternatives, PackWriter may pack the same
object twice (once in the thin section, again in the cached base
pack). This may cause the command line C version to go into an
infinite loop when repacking the resulting repository, as it may see
a delta chain cycle with one of those duplicate copies of the object.

Previously the DfsReader tried to avoid looking at packs that it
might not care about, but this is insufficient, as all versions
must be considered during pack generation.

Change-Id: Ibf4a3e8ea5c42aef16404ffc42a5781edd97b18e