Clarify if zero-duration spans are valid?

[pmm-sumo]

What are you trying to achieve?

Should we provide clarification if zero-duration spans are valid? Basing on opentelemetry-proto those seem to be OK, however I wasn't able to find this mentioned anywhere in the tracing specs.

If such spans are acceptable, it is not clear how they should be interpreted (not sure if on any currently existing platform monotonic clock could measure less than 1 nanosecond)

Also, perhaps this is obvious but I haven't found a piece of specification mentioning that end timestamp must be greater or equal to start timestamp (like in the proto comments)

Overall, this is perhaps a rather minor issue but saw some confusion over that so I am brining it up here for discussion

[yurishkuro]

sometimes zero-length span is a modeling choice to compensate for span model's inability to express causality directly between events