Support throughput benchmarking for mllama with vision input

[yisonzhu]

Before, the benchmark_through.py did not support multi-modal data for most of the VLMs like mllama. The purpose of this PR is to facilitate our throughput test for mllama on HPU.
Only for internal test, not aiming for upstream.

Usage

We can now test with such commands:

export VLLM_DECODE_BLOCK_BUCKET_MAX=384
export VLLM_PROMPT_SEQ_BUCKET_MAX=128
python benchmark_throughput.py --model=meta-llama/Llama-3.2-11B-Vision-Instruct --max-model-len=4096 --input-len=13 --output-len=40 --num-prompts=32 --max-num-seqs=4 --mm-data