Make TransformerEncoderLayer layer norm names more descriptive

Summary:
I added an upgrade_state_dict function so that loading old models will still work

layer_norms[0] --> self_attn_layer_norm
layer_norms[1] --> final_layer_norm

Reviewed By: pipibjc

Differential Revision: D14689849

fbshipit-source-id: b2809262c11fe9d083e571fa31044798aefd48ce