Fix locking for getmempoolentry RPC command

• Fix getmempoolentry by adding WRITELOCK

  We are updating tx chain state and so need a write lock rather
  than a read lock. Somehow this was not getting picked up by
  the assertwritelockheld() on L422 in txmempool.cpp

• Add second lookup in getmempoolentry

  By using a second lookup in getmempoolentry, if and only if the
  initial lookup shows that the entry is dirty, then we can in
  almost all cases only require a READLOCK and one lookup but
  if the transactions chain is longer that 500 then we can still
  update it correctly and return the correct state.