[Performance] Faster lock_/unclock_ when sub-tds are already locked #2726

Job	Run time
build-wheel-linux (3.8, cp38-cp38, --extra-index-url https://download.pytorch.org/whl/cpu, "['cpu...	1m 6s
build-wheel-windows (3.8, 3.8)	1m 52s
build-wheel-windows (3.9, 3.9)	1m 50s
build-wheel-windows (3.10, 3.10.3)	3m 27s
build-wheel-linux (3.9, cp39-cp39, --extra-index-url https://download.pytorch.org/whl/cpu, "['cpu...	1m 5s
build-wheel-windows (3.11, 3.11)	2m 53s
build-wheel-linux (3.10, cp310-cp310, --extra-index-url https://download.pytorch.org/whl/cpu, "['...	1m 2s
build-wheel-linux (3.11, cp311-cp311, --extra-index-url https://download.pytorch.org/whl/cpu, "['...	1m 9s
test-wheel (linux, ubuntu-20.04, 3.8)	54s
test-wheel (linux, ubuntu-20.04, 3.9)	50s
test-wheel (linux, ubuntu-20.04, 3.10)	56s
test-wheel (linux, ubuntu-20.04, 3.11)	1m 0s
test-wheel-windows (3.8)	2m 16s
test-wheel-windows (3.9)	3m 13s
test-wheel-windows (3.10)	3m 0s
test-wheel-windows (3.11)	2m 24s
	28m 57s

Provide feedback