Envoy AI Gateway internal/tracing/ - per-endpoint OTel tracers with W3C autopropagation

Read at main 2026-10-05. tracing.go builds a sdktrace.TracerProvider configured entirely from OTEL_* environment variables (OTEL_SDK_DISABLED, OTEL_TRACES_EXPORTER, OTEL_EXPORTER_OTLP_ENDPOINT, OTEL_EXPORTER_OTLP_TRACES_ENDPOINT, OTEL_BSP_SCHEDULE_DELAY, OTEL_SERVICE_NAME, OTEL_RESOURCE_ATTRIBUTES) and installs propagators via autoprop.NewTextMapPropagator(), i.e. OTEL_PROPAGATORS. One typed tracer per endpoint (ChatCompletion, Completion, Embeddings, ImageGeneration, Responses, Speech, Transcription, Translation, Rerank, SystemOne, Message, Tokenize, ResponsesInputTokens, CountTokens, MCP). Two semantic-convention families are implemented side by side: openinference/ and otelgenai/. span.go buffers streaming chunks before ending a span so a streamed response is recorded as one span.