This change simplifies handling of access entities in access storages.

This test triggers heartbeat failrues which leads to
REBALANCE_IN_PROGRESS which causes hang.

After first INSERT for disk2 there should not be created any per-shard
directory, so find(1) should report an error, like:

  find: '/disk2/data/test/dist_foo/default@127%2E0%2E0%2E2:9000': No such file or directory"

But sometimes output can be fixed, and output of wc(1) will goes first
and python's int() will parse it and not fail, but if find(1) stderr
will goes first the int() will fail to parse.

And here is an example of such mixing:

  $ docker run --name alpine --rm -it alpine top
  $ docker exec alpine sh -c 'echo foo >&2 | wc -c'
  foo
  0
  $ docker exec alpine sh -c 'echo foo >&2 | wc -c'
  0
  foo

Add setting  which restrict replica to merge parts by itself and always prefer dowloading from other replicas.