Opprotunistically prune run condition spans

## What problem does this solve or what need does it fill?
A span for profiling has non-insignificant overhead (#4892) and a good chunk of run conditions don't take very long to run. The construction of these spans is a few orders of magnitude greater than the time it takes to run the condition itself. This is skewing profiling measurements of the executor when there are a lot of these conditions.

## What solution would you like?
Opportunistically avoid making these spans when we know that the time spent in the condition is low.

## What alternative(s) have you considered?

1. Leave it be. Accept the overhead and factor it into our evaluations of performance changes.
2. Omit them entirely.
3. Put them behind a feature flag.