* log best.pt metrics at train end

* update

* Update __init__.py

* flush callbacks when using evolve

* remember batch size on resuming

* Update train.py
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>