Adding past inputs to MultiHeadAttention parser for supporting new models

[urpetkov-amd]

Motivation

This PR has been created from this original PR: #4637
Motivation for this is support more input in MultiHeadAttention parser in order to support new VLM models.

Technical Details

Past key, past value and past_sequence_length have been added as new inputs that MultiHeadAttention supports.
Tests have been added for these new inputs.

Changelog Category

Add a CHANGELOG.md entry for any option other than Not Applicable

• • Added: New functionality.
• • Changed: Changes to existing functionality.
• • Removed: Functionality or support that has been removed. (Compared to a previous release)
• • Optimized: Component performance that has been optimized or improved.
• • Resolved Issues: Known issues from a previous version that have been resolved.
• • Not Applicable: This PR is not to be included in the changelog.

[codecov]

Codecov Report

❌ Patch coverage is 97.64706% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/onnx/parse_multi_head_attention.cpp	97.65%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #4703      +/-   ##
===========================================
+ Coverage    92.32%   92.34%   +0.01%     
===========================================
  Files          583      583              
  Lines        29334    29412      +78     
===========================================
+ Hits         27082    27158      +76     
- Misses        2252     2254       +2     

Files with missing lines	Coverage Δ
src/onnx/parse_multi_head_attention.cpp	98.25% <97.65%> (-0.20%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[TedThemistokleous]

Overall looks good.

• Good comments explaining logic (gen_onnx test cases, early returns)

Just add the following

• Leverage std::optional for capturing the argument after you've validated that all past input are good. This will simplify your checks and not rely on size of args
• Add a small blub for your verify tests on how you generated the gold data in a comment box. Good for reproduction if we somehow see a regression with some other MIGraphX change.

[urpetkov-amd]

@TedThemistokleous
Thanks. Changes have been added.

[urpetkov-amd]

@TedThemistokleous @causten

Can we merge this if everything is okay now? Thanks