[optimistic] Per-instance waitgroup inaccurately reflects status of optimistic processing

**Describe the bug**
As pointed out by Auston and Max of aestus relay, the current `waitgroup` design for handling optimistic block processing only works on a per-instance basis. In order to work as intended, some cross-process synchronization step is needed. Probably the best way to do this is through redis.

**Expected behavior**
The waitgroup is incremented up receipt of each optimistic builder submission: https://github.com/flashbots/mev-boost-relay/blob/c52858a6687ae538685b753a3c8323bf9b7d0e5d/services/api/service.go#L609

It is then waited on in a few places to try to ensure there is no outstanding optimistic blocks being processed: https://github.com/flashbots/mev-boost-relay/blob/c52858a6687ae538685b753a3c8323bf9b7d0e5d/services/api/service.go#L1304

Since there are multiple builder API and proposer API instances, the per-instance waitgroup isn't sufficient to ensure that all the optimistic blocks have been processed.

**Additional context**
Auston has explored changes to redis to handle this coordination: https://github.com/austonst/mev-boost-relay/commit/f52c9771b32a3d266db8b87b133f9d94b8a6179d

Read services/api/service.go at the linked submission and wait locations to understand how the per-instance waitgroup is updated and consumed. Review the linked Redis exploration for coordination ideas; done means waits account for optimistic blocks across all builder and proposer API instances.

Written by the indexing model from the issue text.