newFixedThreadPoolContext/newSingleThreadContext can OOM on long timeouts

[elizarov]

Consider the following code:

fun main(args: Array<String>) = runBlocking {
    val scope = CoroutineScope(newSingleThreadContext("test"))
    while (true) {
        val job = scope.launch(start = CoroutineStart.ATOMIC) {
            delay(Long.MAX_VALUE)
        }
        job.cancelAndJoin()
    }
}

It creates newSingleThreadContext, then launches and cancels lots of coroutines that do delay(Long.MAX_VALUE) there. This code will end with OutOfMemoryError, because newSingleThreadContext uses ScheduleTheadPoolExecutor but cancelling scheduled tasks in it does not really work (memory is not really released until the trigger time). We'll need to rewrite newSingleThreadContext on top of a better thread-pool or use system-wide (default) delay implementation (aka DefaultExecutor) that does not suffer from this problem. We shall consider this problem in scope of #261.