Fix OTLP exporter Tokio runtime usage

[fmassot]

Summary

• Wrap OTLP trace, log, and metric exporters in a small stable-trait adapter that owns a Tokio runtime for export calls.
• Keep OpenTelemetry's stable batch processors instead of enabling experimental_async_runtime, experimental_trace_batch_span_processor_with_async_runtime, or experimental_logs_batch_log_processor_with_async_runtime.
• Make telemetry shutdown best-effort across trace/log/metric providers and run CLI telemetry shutdown on a plain OS thread to avoid dropping Tokio runtimes from a Tokio worker.

Root Cause

OpenTelemetry's stable batch processors export from dedicated OS threads. The OTLP exporters can use Tokio-backed HTTP/gRPC paths internally, which need a Tokio reactor. Without one, the batch processor thread can panic with there is no reactor running, must be called from the context of a Tokio 1.x runtime.

Validation

• cargo check -p quickwit-telemetry-exporters
• cargo check -p quickwit-cli