Fix vLLM Gemma4 support by adding v2 module fallback and patch target fallback

[mahendrarathore1742]

• Added support for vLLM v2 module layout fallback in vLLM integration patches.
• Updated PatchManager to accept multiple target modules and try alternate import candidates before failing.
• Added fallback-aware imports for vllm.v1.* and vllm.v2.* in patches.py and nixl_compat.py.
• Added regression coverage in test_vllm_v2_support.py to verify vLLM import fallback and patch manager module selection.
• This resolves Gemma4 model loading issues caused by newer vLLM engine path changes.