Doublecheck GRPC Status to Span Status mapping

[lforst]

Followup to #11169

In the logic where we set a span status based on HTTP_STATUS_CODE or RPC_GRPC_STATUS code, there are two inconsistencies with the spec (https://opentelemetry.io/docs/specs/semconv/rpc/grpc/):

• We assume the GRPC status codes are stored as string, while the spec says the GRPC status codes are stored as int.
• We set the status to ERROR for all of the status codes, while the spec says we should set the status to ERROR or UNSET based on whether the span is a client or a server span.

[lforst]

Removed from v9 milestone because I don't think this is breaking.