Use `torch.no_grad()` instead of inference mode in `pytorch_trainer.py` #3688

Job	Run time
unittests	2m 57s
pylint	2m 40s
format	11s
C++ Test Coverage (gets coverage of main branch, currently not main branch because no C++ on main)	2m 55s
Mypy Type Checks	2m 16s
Pre-commit Checks	28s
tidy	2m 58s
Build + Test (C++) (Release, gcc, g++, 11)	1s
Build + Test (C++) (Release, clang, clang++, 14)	1s
Build + Test (C++) (Release, clang, clang++, 17, true)	1s
Build + Test (C++) (Debug, gcc, g++, 11)	1s
Build + Test (C++) (Debug, gcc, g++, 12)	1s
Build + Test (C++) (Debug, clang, clang++, 14)	1s
Build + Test (C++) (Debug, clang, clang++, 17, true)	1s
Build + Test (C++) (clang, clang++, 14, Tsan)	1s
Build + Test (C++) (clang, clang++, 14, Asan)	1s
Comment Coverage Results	1s
integrationtests-release	1s
integrationtests-debug	1s
integrationtests-asan	1s
integrationtests-tsan	1s
dockerized-unittests	1s
	14m 40s

Provide feedback