Feat/optimize rpc

[konradkonrad]

Select optimizations to reduce RPC call count, based on analysis from #717:

S2 optimization

• keyperimpl/shutterservice/eventprocessor.go — new interface
  - FilterCriteria struct carries Addresses []common.Address and Topics0 []common.Hash (empty
  Topics0 means "any topic" for that processor).
  - EventProcessor interface replaces the old FetchEvents(ctx, start, end) with FilterCriteria(ctx, start, end) + ParseEvents(ctx, start, end, logs).

• keyperimpl/shutterservice/multieventsyncer.go — combined fetch
  - gatherFilterCriteria asks each processor for its criteria, builds the unioned address list and topic[0] list, captures per-processor scopes for routing.
  - syncRange now issues at most one FilterLogs(addresses, [topic0Union]) per range (skipped entirely if no processor wants any addresses), then routes the matching subset of logs to each processor's ParseEvents. If any processor declines to constrain topic[0], the combined query drops topic filtering.

• keyperimpl/shutterservice/eventtriggerregisteredprocessor.go
  - Constructor now takes the contract address and parses the bindings' ABI to extract the EventTriggerRegistered topic. Returns an error from construction.
  - FilterCriteria returns {[address], [topic]}.
  - ParseEvents calls Contract.ParseEventTriggerRegistered(log) (abigen helper) per log.

• keyperimpl/shutterservice/triggerprocessor.go
  - New activeTriggers helper reads + parses active trigger definitions from the DB.
  - FilterCriteria unions the addresses and topic[0] hashes across all active triggers; sets anyTopic when a trigger leaves topic[0] unconstrained.
  - ParseEvents iterates each trigger over the pre-filtered logs and emits TriggerEvents for ones that pass def.Match and aren't expired.

• keyperimpl/shutterservice/keyper.go
  - Updated NewEventTriggerRegisteredEventProcessor call to pass the address and handle the new error return.

Steady-state RPC count per block range with N active triggers drops from 1 (FilterEventTriggerRegistered) + N (per-trigger FilterLogs) to 1 combined FilterLogs. The HeaderByNumber(end) is still there (that's S3, not in scope).

Note: there's one DB cost trade-off worth flagging -GetActiveEventTriggerRegisteredEvents is now queried twice per range inside TriggerProcessor (once in FilterCriteria, once in ParseEvents). S8 in the doc covers the cache that would eliminate this.

B4 optimization

• New() precomputes the union of contract addresses, the union of event topic0 hashes, and an (address, topic) -> *EventType dispatch table.
• syncAllInRange() now issues a single FilterLogs covering every registered (address, topic), then dispatches each returned log to its EventType via the precomputed table. Logs that match the query but have no registered handler (cross-product hits) are skipped.
• Removed the per-event-type fan-out (syncSingleInRange) along with the sync.Mutex and golang.org/x/sync/errgroup dependencies.

For the snapshot keyper (the only current caller via chainobserver), this drops per-block eth_getLogs requests from N (one per event type) to 1. Behavior is preserved: the same logs reach Next() in the same block + log-index order.