Refactor Error Handling

[aj-bagwell]

Refactor the error handling:

• Split the Error enum in to recoverable client errors and unrecoverable socket errors.
• Return more errors to client methods instead of crashing the runner
• Improve Scan errors instead of leaving clients hanging when scan fails
• Make run take self as ref so that the socket if errors happen caller can reconnect without invalidating client
• Rely of drop breaking channels instead of explicitly informing client that runner is stopping

[aj-bagwell]

Hi,

Sorry for taking so long to post this follow up pull request with my changes to the error handling.

The first four commits (up to when I split the Error enum ) should be backwards compatible, the last two ("Split Error enum" and "Remove event Channel") are where the big changes are.

I'm posting this more as a starter for discussion to see if you are still interested in the idea. If you don't like it, just say so and it can go straight to /dev/null.

I'm not sure I like how much I am still leaking internal details e.g. in ParseError or SocketError::ClientChannelClosed.

I am also not sure about how to treat SelectResult as returning Ok(SelectResult::InvalidNetworkId) feels a bit like "Task failed successfully". On one hand I am tempted to a RunnerNotRunning or something to SelectResult then fn select_network would not need to wrap it in a result. On the other I am tempted to move PendingSelect InvalidNetworkId and Timeout into ClientError. InvalidNetworkId should apply to more things (e.g set the netwokr ssid) but wpa supplicant just returns "FAIL"

And finally there is the broader issue of timeouts, my gut feel is that there are still ways in which the clients can hang indefinitely, and every request should have a similar timeout mechanism to select_network.