fix: use the correct dispatches for device overloads #3820

Job	Run time
downgrade	11m 39s
test (1.10, ubuntu-latest, core_layers)	1s
test (1.10, ubuntu-latest, normalize_layers)	6m 45s
test (1.10, ubuntu-latest, recurrent_layers)	1s
test (1.10, ubuntu-latest, autodiff)	1s
test (1.10, ubuntu-latest, misc)	1s
test (1.10, ubuntu-latest, reactant)	1s
test (1, ubuntu-latest, core_layers)	1s
test (1, ubuntu-latest, normalize_layers)	1s
test (1, ubuntu-latest, recurrent_layers)	1s
test (1, ubuntu-latest, autodiff)	1s
test (1, ubuntu-latest, misc)	1s
test (1, ubuntu-latest, reactant)	1s
test (1, macos-latest, all)	18m 22s
test (1, windows-latest, all)	1s
	36m 58s

Provide feedback