Implements native OTLP/gRPC output enabling direct gNMIC → OTEL Collector
integration without intermediate components (NATS/Prometheus/Kafka).

Features:
- gRPC transport with TLS support
- Automatic metric type detection (Counter/Gauge based on path heuristics)
- Proper gNMI path → OTLP metric name conversion (slash/hyphen to underscore)
- Configurable subscription name prepending for vendor-specific prefixes
- Event.Values iteration supporting any value key structure
- Comprehensive validation of OTLP message structure
- Response PartialSuccess checking for rejection detection
- Configurable batching with worker pools
- Retry logic with exponential backoff
- Prometheus metrics for observability

Configuration options:
- endpoint: OTLP collector endpoint (required)
- protocol: "grpc" or "http" (default: grpc)
- timeout: Request timeout (default: 10s)
- batch-size: Metrics per batch (default: 1000)
- interval: Max time before sending batch (default: 5s)
- num-workers: Worker pool size (default: 1)
- max-retries: Retry attempts (default: 3)
- append-subscription-name: Prepend subscription name to metrics (default: false)
- strings-as-attributes: Convert string values to gauge with attribute (default: false)
- metric-prefix: Global prefix for all metrics (optional)
- resource-attributes: Static resource attributes (optional)
- tls: TLS configuration (optional)

Example configuration:
outputs:
  otlp:
    type: otlp
    endpoint: otel-collector:4317
    protocol: grpc
    batch-size: 1000
    num-workers: 2
    append-subscription-name: true
    strings-as-attributes: true
    resource-attributes:
      telemetry.source: "gnmi"

Metric naming:
- [prefix_][subscription_]path_with_underscores
- Example: nvos_interfaces_interface_state_counters_in_octets
- Complies with OTLP naming conventions (a-z, 0-9, _, .)

Updated all.go to have distinct copyright blocks for Nokia and NVIDIA
rather than combining them, as requested by the maintainers.

Updates the OTLP output to use grpc.NewClient instead of the deprecated
grpc.DialContext. This change ensures the output can initialize successfully
even when the OTLP collector endpoint is unreachable at startup.

Key changes:
- Use grpc.NewClient which creates client without dialing
- Connection now happens lazily on first RPC instead of during Init
- Allows OTLP output to start and retry when endpoint becomes available
- Remove unnecessary timeout context from initialization

Adds comprehensive unit tests to verify:
- Init succeeds with unreachable endpoints
- Connection established on first RPC call
- Reconnection works when endpoint becomes available
- Normal operation with reachable endpoints