feat: reconsider fallback design

[ncitron]

Right now the fallbacks function somewhat strangely. We have these checkpoint providers but we only use them when the initial sync fails. This doesn't capture the case when the initial checkpoint is too old and will cause the sync to be too slow. The current way around this is to use --strict-checkpoint-age and --load-external-fallback/--fallback together.

I don't know but this whole design seems odd. I'm open to suggestions on better ways to incorporate the checkpoint providers into Helios in a more sensible way.