Emit matching_timeout outcome for Nexus matching dispatch errors

[bergundy]

When the matching client returns an error from DispatchNexusTask in StartOperation and CancelOperation, tag the metrics with outcome=matching_timeout instead of falling through to the default internal_error outcome. This lets us distinguish failures originating from the matching dispatch (typically transient timeouts) from other internal errors, which are considered more severe.

[chrsmith]

This is totally fine as-is. But the tags seem kinda like magic numbers. Does it make sense to expose values from some shared metrics package, like:

...WithTags(metrics.MatchingTimeoutTag)

[bergundy]

This is totally fine as-is. But the tags seem kinda like magic numbers. Does it make sense to expose values from some shared metrics package, like:

...WithTags(metrics.MatchingTimeoutTag)

I would be okay with following up with that in a different PR. I'm personally not a fan of defining constants for things that have limited reuse, I prefer not to have to jump around through the code to see the full picture but that's just a personal preference.