Check the size of objects being read from the loose odb backend and
reject those that would not fit in memory with an error message that
reflects the actual problem, instead of error'ing later with an
unintuitive error message regarding truncation or invalid hashes.