## What changes were proposed in this pull request? - Fixes the issue with the frameworkId being recovered by checkpointed data overwriting the one sent by the dipatcher. - Keeps submission driver id as the only index for all data structures in the dispatcher. Allocates a different task id per driver retry to satisfy the mesos requirements. Check the relevant ticket for the details on that. ## How was this patch tested? Manually tested this with DC/OS 1.10. Launched a streaming job with checkpointing to hdfs, made the driver fail several times and observed behavior:          Author: Stavros Kontopoulos <st.kontopoulos@gmail.com> Closes #19374 from skonto/fix_retry.
