cli: Prune gather branches to speed up "result merge"

The "gather" operation which is part of "result merge" can be quite
expensive due to needing to hash objects to check if they are already in
the set. Previously we were massively over-gathering, since we would try
to gather the same entity multiple times.

Consider a set of results for the same benchmark on the same sut: For
every result we will end up gathering the same benchmark entity, sut,
nodes, etc. We can avoid this by marking entities that are already
gathered, then we can avoid the extra pointless recursion.

With this optimisation and the previous one combined, a merge that
previously took over 2 mins, is down to 30 seconds.

Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>