fix(chainwatch): Make processor failures independent of each other

[placer14]

Chainwatch processor works in batches, collecting the deltas for each batch and shipping them off to sub-processing handlers which work through persisting the deltas to the DB. When any part of sub-processing fails, the errgroup cancels the whole batch. Instead, let's make each sub-process handler persist best-effort since they don't have schema dependencies between one another and log the failure. This should allow us to capture much more of our data despite internal state inconsistencies arising and making the whole thing bail.

Also, reduce the batch size. Smaller batches means a less efficient catchup when chainwatch lags, but the trade off is smaller gaps in data on failures.