fix: issue 675：goroutine leak when connectionUp(true) return error

[kiqi007]

bug fix: #675

• Skip the check for Disconnect when the connection is being established.
• Treat this scenario (Disconnect during connection) as if the Disconnect occurred after connectionUp(true) completed.

[MattBrittan]

So, in summary, this PR will change things so that:

if a connection is in progress when Disconnect is called then the connection attempt will be completed (including calling OnConnect and spinning up goroutines) and then the Disconnect operation will proceed after this (so, effectively a normal disconnect). In this case connectionStatus.Disconnecting() will wait until the connection process has completed before returning (and allowing the disconnection process to proceed). startCommsWorkers maintains a lock on connMu throughout the process, so stopCommsWorkers will be blocked until it completes.

My feeling is that this should work (but it's quite difficult to trace all possibilities). However I'd prefer to see the old code removed rather than being commented out (it will be available in the git history) and would like to see a comment something like the above (i.e. we will allow the connection to complete before the Disconnect is allowed to proceed).

I think I'm OK to accept this but if you are able to add something like the following I'd appreciate it. Adding a test for this would be absolutely fantastic (but I'm thinking it might be quite difficult to write).

[kiqi007]

Of course, I have followed your suggestions.
For testing, I simulated this scenario locally using the method I previously used to debug the issue, and it works.

[MattBrittan]

Thanks.

For testing, I simulated this scenario locally using the method I previously used to debug the issue, and it works.

This comment was more about adding a test (that does not pass with the old code but does with the new). The main purpose of this is to help prevent regressions (the state management stuff is pretty complex and Disconnect is not used all that often so it would be east to reintroduce the issue). Anyway I'll accept this PR now and if you do have a test that could be added later.