Bechmarks: Model Benchmark - Add support of megatron lm deepseek v2 lite model training for cuda gpus

[yukirora]

Description
Add support of megatron lm deepseek v2 lite model training for cuda gpus.

Major Revision

• Upgrade Megatron-lm submodule to support mock-data and later models
• Add support for unknown args parse in megatron-gpt benchmark to support new args in new megatron-lm
• Register megatron-deepseek-v2-lite benchmark for cuda

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.73%. Comparing base (c65ae56) to head (5609a09).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #765      +/-   ##
==========================================
+ Coverage   85.69%   85.73%   +0.03%     
==========================================
  Files         102      102              
  Lines        7691     7709      +18     
==========================================
+ Hits         6591     6609      +18     
  Misses       1100     1100              

Flag	Coverage Δ
cpu-python3.10-unit-test	71.02% <100.00%> (+0.07%)	⬆️
cpu-python3.12-unit-test	71.02% <100.00%> (+0.06%)	⬆️
cpu-python3.7-unit-test	70.47% <100.00%> (+0.07%)	⬆️
cuda-unit-test	83.64% <100.00%> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.