Fix OpenTelemetry runtime and shutdown

[kylewlacy]

This PR fixes some issues around OpenTelemetry following #140: namely, trying to use BRIOCHE_ENABLE_OTEL=1 would cause a deadlock, and even after fixing that, events wouldn't be flushed properly. There were two fixes involved:

1. Enable Tokio batched OpenTelemetry exporter. This fixes a panic that was silently being ignored, caused by some HTTP code trying to access the Tokio runtime on a non-Tokio thread
2. Update OpenTelemetry shutdown code. opentelemetry::global::shutdown_tracer_provider() no longer seems to do it. Now, we explicitly shut down the tracer provider directly.