Span.End should record panic

[pellared]

Why

document that End should handle recording exceptions for panics.

I think that documenting that span.End should handle recording exceptions for panics is a good idea.

In my opinion, this is how we implement recording an exception in the most possible user-friendly way.

Originally posted by @pellared in #7254

What

Document that

opentelemetry-go/trace/span.go

Lines 28 to 32 in cba6502

	// End completes the Span. The Span is considered complete and ready to be
	// delivered through the rest of the telemetry pipeline after this method
	// is called. Therefore, updates to the Span are not allowed after this
	// method has been called.
	End(options ...SpanEndOption)

should also record an unhandled panic and set status to Error.

In order to follow https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/trace/exceptions.md#recording-an-exception, then here

opentelemetry-go/sdk/trace/span.go

Lines 465 to 482 in bc531cb

	if recovered := recover(); recovered != nil {
	// Record but don't stop the panic.
	defer panic(recovered)
	opts := []trace.EventOption{
	trace.WithAttributes(
	semconv.ExceptionType(typeStr(recovered)),
	semconv.ExceptionMessage(fmt.Sprint(recovered)),
	),
	}
	if config.StackTrace() {
	opts = append(opts, trace.WithAttributes(
	semconv.ExceptionStacktrace(recordStackTrace()),
	))
	}
	s.addEvent(semconv.ExceptionEventName, opts...)
	}

we also should

• set span status to Error with a description
• set error.type attribute (semconv: Add ErrorType function #6904)