We shouldn't lose any data if writecache format is updated.
Possibilities:

1. External command to flush all writecache-only items to the blobstor.
2. Same but via the control RPC (executed on the old node version before the update)
3. Migration (can work here, because the format is very simple)