[Test-Proxy] Make test recorder more efficient when recording LROs

[m-nash]

Library name

Azure.Core.TestFramework

Please describe the feature.

Right now when an LRO is recorded in the worst case where the service doesn't send back a Retry-After header the default will be to query for status every second. In management plane quite a few of our test cases take 15-60 minutes to record which results in an extremely large file which adds to test playback execution time since the recorder will need to find and deserialize all of those.

The enhancement would be to give the recorder context that it is recording an LRO and only log

1. Initial request
2. First retry after request
3. last retry before final response
4. final response

With this set you can properly test an LRO scenario as all other recordings will be identical to Azure/azure-sdk-for-net#2. We could also potentially look at only recording if current != last just in case somewhere in the middle we get a slightly different response (I don't think this is possible).

[azure-sdk]

Label prediction was below confidence level 0.6 for Model:ServiceLabels: 'Cognitive - Form Recognizer:0.26791137,Service Bus:0.17913784,Storage:0.07234221'

[JoshLove-msft]

This should probably be a Test Proxy enhancement. One idea we had was to do a pass at the end of recording an LRO test where we remove duplicate request/response pairs. The mgmt team already has a script that they use to do this cleanup.