[quant][graphmode] Dynamic quantInsert observers for module output (#39458)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/39458
Previously if we had a CallMethod followed by a CallFunction, we didn't check for observers at output of CallMethod since it was handled separately.
This change makes it default to check the outputs of all nodes to identify values that need observers
Test Plan:
python test/test_quantization.py test_dynamic_shared_weights
Imported from OSS
Differential Revision: D21872939
fbshipit-source-id: 08dd8b7ddf73ef2cc26ebcf4ceb2f222c4559ab3
