Chain DB: Profile memory usage of chain selection

Chain selection (current WIP PR: input-output-hk/ouroboros-network#617) involves (mainly) computing new ledger states. We need to profile this to make sure that we don't end up with huge memory usage due to, for example, multiple copies of the ledger state.

We have already verified that when doing chain selection with the Byron ledger, without any forks, the ledger snapshots don't require much additional memory. We should verify that this is also fine for the Shelley, when there are much more forks.