Feat: add command-line arguments for backend parameters

[SyedaAnshrahGillani]

This pull request enhances the flexibility of the gpt_oss/generate.py script
by replacing hardcoded backend parameters with configurable command-line
arguments.

Previously, the triton and vllm backends had fixed values for context and
tensor_parallel_size, respectively. This made it difficult to adapt the
script to different models or hardware configurations without modifying the
source code.

This PR introduces two new arguments:

• --context-length: Allows customization of the context length for the triton
  backend (defaults to 4096).
• --tensor-parallel-size: Allows customization of the tensor parallel size for
  the vllm backend (defaults to 2).

Additionally, the variable decoded_token has been renamed to token_text for
improved clarity.

These changes make the generation script more versatile and user-friendly,
allowing for easier experimentation with different backend settings.