Call to NVIDIA's NvDsSRDestroy function blocks indefinately if Pipeline stoped while recording in progress. (NVIDIA Bug)

[rjhowell44]

This was first reported back in https://forums.developer.nvidia.com/t/end-of-stream-during-smart-record-throwing-error-runusercallback-199/188817/3

I had thought that the workaround was to try ensure that the recording is stopped before the pipeline is stopped.... but the only works some of the time.

The workaround now is to NOT destroy the context when the Pipeline is stopped and unlinked. This will leak data of course. Won't be a big deal as this happens the application is closing for most users. It is only when a Pipeline is stopped and play repeatedly that this becomes an issue.