Replace panics in dynamic location handling with safe fallback

[utpilla]

Summary

EventFormat::get_data_with_offset_direct had two todo!() arms for LocationType::DynRelative and LocationType::DynAbsolute that panic the process if hit. This change replaces them with a single combined arm that returns EMPTY, matching the behavior of every other failure path in the same function.

Motivation

These variants are produced by Linux tracefs (parsing __rel_loc / __dyn_loc markers) and by the cross-platform ScriptEvent::append_field helper. Whether any current caller actually feeds such a field into get_data_with_offset_direct is not obvious from a single review pass. For example: open-telemetry/otel-arrow#3187 (comment)

A todo!() here forces reviewer to either:

• ask for defensive scaffolding (catch_unwind, etc.) around callers, or
• chase the call graph to prove the branch is unreachable.

Both are wasted effort for a function whose every other failure mode already returns EMPTY. todo!() is a "not implemented yet" marker; using it in shipped code as a stand-in for "shouldn't happen" promotes a missing feature to a potential process-killing panic and adds review friction.

[beaubelgrave]

We need to handle these, as the data closure generation calls into get_field_data_closure() which eventually leads here. get_field_data_closure() should be updated to handle these cases as well as this function.

The todo() here is to track that work is yet to be done. We should either do the work here or keep the todo() since it's tracking that left over work.

[utpilla]

Closing in favor of #293