change profiling based partitoner's last_segment_latency to slowest_segment_latency

[HanChangHun]

Changed profile-based partitioner's last_segment_latency to slowest_segment_latency.

Actually it does not guarantee gap between slowest_segment_latency and fastest_segment_latency lower than diff_threshold_ns. But when using this code, I can see that the difference was reduced.

I tried to use a different method, but I didn't converge.

And I tested by utilizing only one TPU. So I couldn't proceed with the experiment on more than 2 TPUs.
I'm using M.2 Accelerator with Dual Edge TPU, but I can't catch two TPUs... So my test can be inaccurate.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.