[Bugfix] userspace convertor: Fix race condition on error

[estebanreyl]

I recently discovered that there is a race condition on the user space convertor (during build) when an error occurs which can lead to a Panic. To Summarize:

0. All the goroutines are scheduled
1. Error occurs on one of the subroutines (To repro for example I returned an error 30% of the time in the Download function)
2. The error is sent into the retErr channel.
3. The retErr goroutine receives the retErr and cancels the context.
4. The upload goroutines which are all waiting for either ctx.Done() or their respective conversion channel all end.
5. The main build goroutine moves past uploaded.Wait() which results in ending the function as an error was returned. In turn this calls the deferred close on errCh, note that that at this point not all goroutines are guaranteed to be done.
6. A goroutine for example one of the download ones reaches an error state (because context cancelled most likely)
7. This goroutine will now try to sendToChannel errCh or check for ctx.Done and since selects have no order guarantee in Golang however either can happen, if its the send on errCh this will result in a panic as the channel is already closed.

To fix it one way was to add more values to the uploaded waitgroup and make sure all goroutines finish but instead I opted to use sync/errgroup https://pkg.go.dev/golang.org/x/sync/errgroup which seems built for this exact kind of scenario.

Any feedback is appreciated.

[WaberZhuang]

LGTM. Thanks a lot for this bugfix! sync/errgroup looks quite suitable for this scenario.

[liulanzheng]

LGTM