Consider adding tools to automatically trace step inputs/outputs

[apoliakov]

A user writes:

In many cases, the inputs to a Step and the output from that Step are valuable as part of the trace.
When I enable OTEL for DBOS it will log spans for Workflow and Steps - but to capture the input/outputs I will need to write all the boiler plate code that does that (get the DBOS span, log the input, finally get it again, log the output) all of that can be simply added to the infra as an extension to the Step decorator to request logging the input and output - again this is relevant in the context of Langfuse and other tracing platforms that have the concept of "Input" and "Output" - this is not (AFAIK) part of the OTEL tracing basic functionality.

[yoadsn]

A small input on that, the mapping between runtime values and eventual OTEL attributes is always a decision of the user, of course. I just wonder how can it be easier to plug into DBOS existing OTEL support to be able to do something like the input/output capture which is very valuable in the context of AI Agents. But this mechanism (once defined) could be used for any other need.
What I have done today to get around this is a decorator that is applied "below" the "@step" decorator (so it is wrapped in a DBOS span), get the current span, applies what ever attribute mapping I need, change the name of the span, and does the same at the end of the wrapped call. I specifically target langfuse in that wrapper.