The console randomly die with an unwrap

[dzamlo]

The console randomly die after some time. Running with RUST_LOG=debug and RUST_BACKTRACE=full produce this log file: console.txt (I suggest printing the log file in a terminal so you get the colors and other formatting)

Error

called `Result::unwrap()` on an `Err` value: SystemTimeError(836.154µs)

Metadata

key	value
version	0.1.1
location	tokio-console/src/state/async_ops.rs:214:61

[zaharidichev]

I think this is due to the fact that essentially before #238 the subscriber was acting like an actor wrt to state updates and these were all driven by messages. So at the time of publishing an update message on the wire it was the case that this message timestamp will be after any other timestamp that was written into the AsyncOp stats. Since these stats are now shared between the aggregator and the ConsoleLayer and guarded by a lock, we can actually observe an Update.now timestamp that is before a Stats.dropped_at timestamp. We should fix that. Two possible ways of doing that:

1. Make sure that when publishing an update we take the now timestamp after everything else has been converted to proto repr.
2. Handle the case in the console where a timestamp can be after the Update.now one.