[CODE] Proper handling of a dictionary miss error

[VladRodionov]

Currently, if dictionary is missing we just return error upstream which is not handled properly.

[VladRodionov]

Will keep it open. Still there is a small chance to read garbled data (race conditions when dictionary is removed on a read path after we check id presence and before we try to decompress). Decompression currently is executed at the very end of a read request.

[VladRodionov]

We need to keep reference to a retired routing table with a time of exact retirement and do lookup of a missing ID if current_time - retire_time < t ( t must be small in order of a several seconds)