fix xgrammar dependency conflict

[do420]

• [CI] Fix mypy pre-commit errors on main ([CI] Fix mypy pre-commit errors on main vllm-project/vllm#36882)
• [Kernel] Add FP8 KV cache support to Triton MLA decode attention ([Kernel] Add FP8 KV cache support to Triton MLA decode attention vllm-project/vllm#34597)
• [Bugfix] fix main branch pre-commit error (1 line change) ([Bugfix] fix main branch pre-commit error (1 line change) vllm-project/vllm#36897)
• [Attention] Use FA4 for MLA prefill ([Attention] Use FA4 for MLA prefill vllm-project/vllm#34732)
• [Bugfix] Fix KeyError in parse_response_input for reasoning items with optional content ([Bugfix] Fix KeyError in parse_response_input for reasoning items with optional content vllm-project/vllm#34499)
• Enhanced paged_attention_v1 function with GPU/CPU compatibility and logging.
• fix: updated xgrammar dependency to >=0.1.30 to resolve conflicts