Add support for HF style Bert

[yaoyu-33]

Is your feature request related to a problem? Please describe.

Post-LN is not supported in MCore transformer at the moment (everything is Pre-LN). In order to convert HF checkpoints, we want to add post-LN style to MCore as well. Post-LN requires an additional LN in the beginning of the network as well.

Describe the solution you'd like

Please refer to NeMo code update here for HF Bert support: #8088