* add head dropout to RepVit as all models have that arg
* default train to non-distilled head output via distilled_training flag (set_distilled_training) so fine-tune works by default w/o distillation script
* camel case naming tweaks to match other models