Currently if we crash in the middle of initial sync, any in-progress index builds will be fully rebuilt before the node starts accepting incoming connections. This is wasted work and keeps the system offline longer, as the moment the node fully finishes startup, it's going to wind up restarting the initial sync and throwing out all data and indexes anyway.

If we can't make them non-restartable, then change the startup code to ignore them during initial sync (since all dbs will be dropped later anyway).