Wire metrics_reporter into the OTLP exporter and flush it on shutdown

[Alan-Marx]

Problem

OtelSetup.build_otlp_exporter constructs OpenTelemetry::Exporter::OTLP::Exporter without passing metrics_reporter::

def build_otlp_exporter(config)
  OpenTelemetry::Exporter::OTLP::Exporter.new(
    endpoint: "#{config.base_url}/api/public/otel/v1/traces",
    headers: build_headers(config.public_key, config.secret_key),
    compression: "gzip"
  )
end

opentelemetry-exporter-otlp's exporter defaults metrics_reporter: nil to its own no-op MetricsReporter when not given one. So a user who configures config.metrics_reporter and exports over OTLP never receives otel.otlp_exporter.failure or otel.otlp_exporter.message.compressed_size - despite SpanProcessor already threading config.metrics_reporter through correctly for the otel.bsp.* metrics. It's an easy gap to miss because everything appears wired: the config option exists, otel.bsp.* metrics work, and there's no error - the two OTLP-exporter-level metrics just silently go nowhere.

Related: Langfuse.shutdown never gives the configured metrics_reporter a chance to flush before process exit, even though it already has explicit shutdown boundaries for the client and tracer provider. For an async reporter (e.g. a DogStatsd client with a background sender thread), the last few metrics from a shutdown-time export can be lost.

Fix

1. build_otlp_exporter now passes metrics_reporter: ResilientMetricsReporter.wrap(config.metrics_reporter, logger: config.logger), matching the existing pattern in SpanProcessor.
2. ResilientMetricsReporter gains an optional #shutdown that forwards to the wrapped reporter only if it responds to :shutdown, rescued the same way as its other three methods.
3. Langfuse.shutdown calls it after OtelSetup.shutdown.

Testing

• spec/langfuse/otel_setup_spec.rb: a mock-based test on the exporter constructor call, plus a contract test that builds the real tracer provider + real OTLP exporter, stubs the HTTP endpoint to fail with WebMock, and asserts both otel.otlp_exporter.failure and otel.otlp_exporter.message.compressed_size reach a configured reporter through the actual export path (not just a mocked constructor call).
• spec/langfuse/resilient_metrics_reporter_spec.rb / spec/langfuse_spec.rb: cover the new #shutdown passthrough, including the no-shutdown-method and raising-reporter cases.

Both changes are backward compatible - nil metrics_reporter still resolves to the SDK's own no-op reporter, and Langfuse.shutdown is a no-op addition for anyone not using metrics_reporter.

Full suite (bundle exec rspec, bundle exec rubocop) passes locally.

---

Note

Low Risk
Observability wiring and documentation only; tracing export behavior is unchanged aside from forwarding OTLP exporter metrics to an optional reporter the app already owns.

Overview
config.metrics_reporter now feeds both the batch span processor and the default OTLP exporter, so OTLP operational metrics (e.g. otel.otlp_exporter.failure, otel.otlp_exporter.message.compressed_size) reach the same reporter as existing otel.bsp.* metrics instead of silently using OpenTelemetry’s no-op reporter.

OtelSetup wraps the configured reporter once via ResilientMetricsReporter and passes that instance into SpanProcessor and OpenTelemetry::Exporter::OTLP::Exporter. SpanProcessor no longer wraps the reporter itself—it requires an injected metrics_reporter:.

Docs and comments broaden the metrics_reporter contract to cover BSP + OTLP, state that Langfuse does not flush, close, or shut down an application-owned reporter, and document shutdown order: Langfuse.shutdown first, then the app flushes/closes backends (e.g. DogStatsD).

Tests cover shared wiring, real OTLP metric emission on failed export, final BSP metrics on shutdown, concurrent shutdown, a UDP integration proving post-shutdown app flush delivers BSP + OTLP metrics, and assertions that Langfuse.shutdown / Langfuse.reset! never call shutdown on the reporter.

^{Reviewed by Cursor Bugbot for commit 854307f. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want higher recall? High effort reviews run extra passes and find more bugs. A team admin can switch effort levels in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit 36df803. Configure here.}

[kxzk]

@Alan-Marx I updated the branch in commit 854307f.

The refactor passes one resilient metrics_reporter instance to both the batch processor and the default OTLP exporter. Langfuse.shutdown flushes the tracing pipeline but does not flush, close, or shut down the application-owned reporter.

Please verify this sequence with your real reporter integration before we merge:

1. Create and finish a Langfuse observation.
2. Call Langfuse.shutdown.
3. Flush and close the metrics reporter.
4. Confirm that the final otel.bsp.* and otel.otlp_exporter.* metrics reach your metrics backend.
5. Confirm that Langfuse never calls the reporter's shutdown method.

Local verification passed with 1,646 examples, 96.94% coverage, RuboCop, a real UDP lifecycle test, and live Langfuse ingestion plus API readback. Please reply when you have verified the reporter behavior.