Investigate sparse checkout negative glob performance

Looking through some of the NET pipelines (e.g. https://dev.azure.com/azure-sdk/internal/_build/results?buildId=1508998&view=logs&j=a9d5a44a-ac8b-5cae-adb6-1f4d702c5074&t=a13353de-67f0-5622-a4d7-6ae66778033f) I noticed that when sparse checkout includes the negative glob for test recordings, it actually takes much longer than when just including the whole service directory (see the generate matrix job). It's possible the server side CPU cost of the negative glob is not saving time here, at least for service directories with a smaller recording blob size (for example, text analytics vs. storage).

It's possible that sparse checkout should default to inclusive only when checking out a service directory, unless specifically marked as excluded by the service for when blobs are hundreds of MB in size.