ModifiedNodeDoesNotExistException after sending a lot of BGP data and then immediately closing the BGP connection

After a BGP speaker sends a large amount of updates and then closes its session (or ODL closes the session due to missing KeepAlives or something), ModifiedNodeDoesNotExistException may appear. The probability of this exception raises sharply with the amount of data sent to the ODL and the amount of time that passed between the time when the sending is completed and the time when the session is closed. At 100000 paths and closing the session within 10 seconds from the end of sending this probability is something like 99.99999999%.

Steps to reproduce:
1. Configure a BGP speaker with 100000 paths or more and point it to ODL.
2. Start the BGP speaker.
3. Once the BGP speaker is finished speaking, disconnect it (by shutting it down for example).
4. A while after ODL notices the lost session, it will log one or more ModifiedNodeDoesNotExistException with a large bunch of "failed RIB chain" type of errors.

Note: This is NOT a duplicate of YANGTOOLS-383. In the YANGTOOLS-383 sending 5000 paths was enough to see the ODL crashing with this exception and the session died as a direct consequence of the exception. Here the exception occurs as a direct consequence of the session going down.