Clarify that API users SHOULD use valid UTF-8 for exportable data

In https://github.com/open-telemetry/opentelemetry-go/issues/3021, a user reports having spans fail to export because of invalid UTF-8.

I believe the SDK should pass this data through to the receiver, regardless of UTF-8 correctness. I am not sure whether gRPC and the Google protobuf library used by each SDK offers an option to bypass UTF-8 checking, but if so that would be one path forward if we agree.

Another option is to maintain (via sed, e.g.,) copies of the protocol that replace string with bytes; SDKs could use these definitions instead to bypass UTF-8 checking as well.