Connection Migration V2 should detect consecutive write errors on multiple networks and close the session early.

Currently if connection migration v2 is turned on, and a packet write error is triggered on the original network, migration will happen immediately when an alternate network is available. However, if a consecutive packet write error is triggered post migration, the connection will migrate again to the original network, which theoretically could cause a infinite loop, keeping the session open. CMV2 should detect this loop and close session early with write error. 

It would be nice to collect metrics to see how often user encounters more than 2 consecutive write errors during connection migration to understand this scenario. 

Once this is fixed, we can add equivalent tests for QuicStreamFactoryBaseTest::TestMigrationOnMultipleWriteErrors.