[ENHANCEMENT] Supporting Peri-LN architecture (Gemma2 & 3, Olmo2 architecture)

[penpaperkeycode]

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

• Support for the Peri-LN architecture employed by Gemma 2, Gemma 3, and Olmo 2.

Describe the solution you'd like

• Building on the existing Pre-LN code, add an output-LN layer to enable Peri-LN support; a new command-line argument option is also required.

Describe alternatives you've considered

• None

Proposed implementation

• TBU

Additional context

• A prior issue requesting support for the original Gemma architecture was closed as stale.
• Does Megatron has plan to support Gemma？ #707
• Enabling Megatron-LM to support Gemma-family and Olmo 2 architectures.