feat:Add support for Phi-4-mini and Phi-4-MM

[brb-nv]

This MR adds support for Phi-4-mini and Phi-4-multimodal models.

[brb-nv]

/bot run

[kaiyux]

/bot run

[niukuo]

PR_Github #210 [ run ] triggered by Bot

[niukuo]

PR_Github #210 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #222 completed with status: 'SUCCESS'

[brb-nv]

Hi @amukkara @symphonylyh, I'm descoping this MR to have only changes for Phi-4-mini model. Changes for Phi-4-MM will be done in this MR: https://github.com/NVIDIA/TensorRT-LLM/pull/3177

I've addressed all your comments (including those related to the multimodal model). However, current changes for Phi4MM need some overhauling (using Pytorch impl just like Phi3 Vision for encoders instead of TRT engine). I felt separating the two out is cleaner instead of blocking one for the other.

[brb-nv]

/bot run

[tensorrt-cicd]

PR_Github #804 [ run ] triggered by Bot

[brb-nv]

/bot run

[tensorrt-cicd]

PR_Github #805 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #804 [ run ] completed with state ABORTED

[tensorrt-cicd]

PR_Github #805 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #651 completed with status: 'SUCCESS'

[brb-nv]

/bot run

[tensorrt-cicd]

PR_Github #917 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #917 [ run ] completed with state SUCCESS /LLM/main/L0_MergeRequest_PR pipeline #724 completed with status: 'SUCCESS'

[kaiyux]

/bot reuse-pipeline

[tensorrt-cicd]

PR_Github #944 [ reuse-pipeline ] triggered by Bot

[tensorrt-cicd]

PR_Github #944 [ reuse-pipeline ] completed with state SUCCESS Reusing PR_Github #917 for commit 179d06d