We should discuss a pattern for very long-running operations (VLROs)

[JeffreyRichter]

I think we do NOT want to recommend the LRO pattern for very long-running operations (VLRO) because the client will effectively hang if polling to completion to make forward progress. We need a time when an LRO transitions to a VLRO; for example, I think we'd all agree that 1 hour is a VLRO and that clients should not poll for 1 hour. I suspect we'd handle this like Johan's Build progress example that he sometimes uses. In which case, we already have the pattern so e just need concrete guidance indicating when to switch from our LRO pattern to our existing GET pattern.