Improve scale writer creation based on producer buffer

[wenleix]

We observe two cases in production cause current scale writer heuristics
not able to scale:

1. When there is skew on the producer side and more than half of the
   producer buffer is not overutilized.
2. When grouped execution is enabled and each bucket doesn't make the
   buffer to be overutilized.

This commit tries to improve the situation by considering overall
producer buffer utilization when deciding scale the writers.

== RELEASE NOTES ==

General Changes
* Improve the scale writer heuristics by considering overall producer buffer utilization. This can be enabled by using the session property `optimized_scale_writer_producer_buffer` and the configuration property `optimized-scale-writer-producer-buffer`.

[arhimondr]

I'm a little bit confused about this loging. I was thinking more about changing the condition to something like if 50% overutilized OR 90% non empty - add more writer. Thoughts?

[wenleix]

@arhimondr : Addressed the comments per offline discussion. Basically now we add more writers if the overall buffer utilization is more than the number of writers.

We don't need to worry about accidentally open too many writers since overall writtenBytes has to be more than writerMinSizeBytes * scheduledNodes.size()

            writtenBytes >= writerMinSizeBytes * scheduledNodes.size()) 

[arhimondr]

This is unused now