Inline TimedRunnable.beforeExecute

[ywangd]

No need to call the method separately as it can be invoked the first time getQueueTimeNanos is called.

Relates: #120488

[elasticsearchmachine]

Pinging @elastic/es-distributed-coordination (Team:Distributed Coordination)

[ywangd]

Just a minor improvement when reading through #120488

[nicktindall]

I'm not sure about this change, if for whatever reason we move the call to getQueueTimeNanos to somewhere else it will return incorrect results?

Also calling it early would erroneously return a time, even if the runnable was still queued?

The only reason for the somewhat convoluted dance here is that the source of createdTimeNanos is not specified so I wanted to avoid exposing it and doing the arithmetic outside TimedRunnable (i.e. we only ever return the deltas between the timestamps and not the timestamps themselves, so there's nothing in the API that indicates whether it's a cached time or not, and hence whether it's safe to compare to a cached/non-cached timestamp)

[ywangd]

if for whatever reason we move the call to getQueueTimeNanos to somewhere else it will return incorrect results?

Also calling it early would erroneously return a time, even if the runnable was still queued?

Both is true. I had assumed there would not be other usages outside TaskExecutionTimeTrackingEsThreadPoolExecutor since TimedRunnable and its relevant methods are package private and used really just inside the Executor. If we move TimedRunnable to be an inner class and change the relevant methods to be private, I think it would make this PR more justified. That said, it is probably not worth the effort. So I am happy to close it.