OTLP adaptor sends trace_id/span_id as hex strings — strict OTLP receivers reject the whole batch ("invalid TraceID length") #3683

Summary

Logflare.Backends.Adaptor.OtlpAdaptor.ProtobufFormatter.build_log_record_fields/1
copies trace_id / span_id values from the log event body verbatim into the
protobuf LogRecord.trace_id / span_id fields:

defp build_log_record_fields({"trace_id", id}) when is_binary(id), do: [trace_id: id]
defp build_log_record_fields({"span_id", id}) when is_binary(id), do: [span_id: id]

Log events carry these as hex strings (32 / 16 characters), but the OTLP spec
defines the protobuf fields as raw bytes of length 16 / 8
(opentelemetry/proto/logs/v1/logs.proto). Spec-conforming receivers reject the
payload; e.g. the OpenTelemetry Collector (otlp receiver, pdata unmarshal) returns:

HTTP 400
{ code: 3, message: "unmarshal: invalid TraceID length" }   (or "invalid SpanID length")

Because the error is at unmarshal time, the entire batch containing one such
event is rejected — with per-second batching this silently drops all logs that
share a batch with any trace-bearing event. Observed in production with a
Supabase log drain (type OTLP, HTTP/Protobuf, mint/1.8.0) pointed at an OTel
Collector v0.155.0 on Cloud Run: batches of Postgres logs (no trace_id) are
accepted (200), batches containing API/edge events are consistently rejected
(400, invalid TraceID length).

Reproduction (against any OTel Collector OTLP HTTP receiver)

An ExportLogsServiceRequest whose LogRecord.trace_id is the 32-byte ASCII
hex string (e.g. "0af7651916cd43dd8448eb211c80319c") → 400 unmarshal: invalid TraceID length. The same payload with the 16 raw bytes
(Base.decode16!(id, case: :lower)) → 200.

Suggested fix

Hex-decode when the value looks like a hex ID, and drop the field (rather than
poison the batch) when it can't be made spec-conformant:

defp build_log_record_fields({"trace_id", id}) when is_binary(id),
  do: id_field(:trace_id, id, 16)

defp build_log_record_fields({"span_id", id}) when is_binary(id),
  do: id_field(:span_id, id, 8)

defp id_field(key, id, size) do
  case Base.decode16(id, case: :mixed) do
    {:ok, bin} when byte_size(bin) == size -> [{key, bin}]
    _ when byte_size(id) == size -> [{key, id}]   # already raw bytes
    _ -> []                                       # unusable — omit, don't poison the batch
  end
end

This is the same bug class as, e.g.,
open-telemetry/opentelemetry-php#1059
(IDs encoded as strings instead of the wire-format bytes).