feat: add OpenFeature track() API for event tracking

[nicklasl]

Summary

Implements the OpenFeature track() API across all three providers (JS, Java, Go) for recording business events tied to feature flag evaluations. This enables A/B test analytics and experimentation metrics via the Confidence events ingestion service.

Data flow

Provider.track(eventName, context, details)
  → construct Event (auto-prefix "eventDefinitions/", merge context+details into payload, set event_time)
  → call wasm_msg_guest_track_event(Event) into WASM
  → EventLogger buffers the Event in a lock-free SegQueue

Provider flush cycle (every ~15s, piggybacked on existing flag log flush)
  → call wasm_msg_guest_flush_events() into WASM
  → EventLogger drains buffer, returns FlushEventsResponse with events list
  → Provider constructs PublishEventsRequest (adds client_secret, sets send_time)
  → Java/Go: gRPC call to EventsService.PublishEvents
  → JS: HTTP POST of serialized protobuf to events.confidence.dev/v1/events:publish

What is included

• Rust core: Lock-free EventLogger using crossbeam_queue::SegQueue
• WASM guest: track_event and flush_events exports
• Proto: Event and FlushEventsResponse added to WASM boundary protos; confidence/events/v1/ service proto vendored for gRPC stubs
• JS provider: track() method, event flush in flush(), HTTP+protobuf sender
• Java provider: track() override, GrpcEventSender using EventsServiceGrpc, event flush on state refresh + shutdown
• Go provider: Track() implementing openfeature.Tracker, GrpcEventSender, event flush piggybacked on log ticker + shutdown

What is not included yet

• Delivery guarantees: Events are currently drained from the WASM buffer unconditionally on flush. If the backend request fails (network error, server error), those events are lost. A future iteration should only drop events from memory after the provider confirms successful delivery from the backend.
• Dedicated event tracking tests: The existing test suites all pass with the new code, but dedicated tests for the track/flush/send cycle (mocking the events endpoint, verifying payload structure, testing flush-on-shutdown) are not yet added.
• Retry/backpressure in event senders: The gRPC event senders (Java/Go) do a single async attempt. Failed sends are logged but not retried. The JS provider has 5 retries via its fetch middleware.

Test plan

• Rust event_logger unit tests (4 tests including concurrency)
• JS provider tests (135 pass)
• Java provider tests (83 pass)
• Go provider tests (all pass)
• Docker full build (docker build .)
• E2E: track events flow through to events backend

🤖 Generated with Claude Code