v0.2.0

Added

• Resilient FlushQueue with bounded memory (5000 cap) and drop-oldest semantics.
• Exponential backoff retries for failed uploads (1s, 4s, 16s), with drop-after-retries warning log.
• sdk_telemetry metadata in CE batch payloads (queue depth, dropped count, flush latency).
• incidentary.fetch() helper for outbound trace header propagation.
• Local pre-arm triggers: slow_success, in_flight_pileup, retry_onset.
• Multi-trigger arbiter: pre-arm enters on one severe or two mild triggers within 10s.
• Enhanced detail capture in PRE_ARMED/INCIDENT modes (method, route, headers, retry metadata, downstream identity).
• Downstream edge key resolution with tiered quality (explicit, route_template, logical_edge, normalized_url).
• Auto-instrumentation integrations: pg, ioredis, grpc, BullMQ, KafkaJS, amqplib.
• Integration registry with auto-detection of installed packages.
• HTTP outbound auto-patching for automatic trace context propagation.
• Ring buffer with captured_before_alert and ring_buffer_seq fields.
• Serverless support with flush-before-exit semantics.

Changed

• Transport returns success/failure to queue orchestrator for retry decisions.