[None][feat] Add XingChen4 model support

[wanghui002]

Background

Onboard XingChen4ForCausalLM — a DeepSeek-V3-style model with MLA + MoE and a multi-Head Hyper-Connection (mHC) residual mixing mechanism — to the TRT-LLM PyTorch backend.

Changes

• Register XingChen4ForCausalLM in the model architecture index, config registry, and pyexecutor config utils (aliased to DeepseekV3Config)
• Add modeling_xingchen4.py: mHC-wrapped decoder layer and dedicated weight loader for hc_fn/hc_base/hc_scale checkpoint tensors
• Route xingchen4 through DeepseekV3MTP for speculative decoding in modeling_speculative.py
• Add XingChen4ReasoningParser (DeepSeek-R1-style with leading <think> strip, delegating to DeepSeekR1Parser or IdentityReasoningParser)
• Add XingChen4ToolParser (<tool_call> / <param_key> block format, streaming-safe)
• Add unit tests for the new reasoning parser (5 new test cases)

Testing

• pre-commit run passes all hooks (isort, yapf, ruff, ruff-format, codespell, DCO check, etc.)
• pytest tests/unittest/llmapi/test_reasoning_parser.py: 213/213 passed
• API stability (signature + docstring): 20/20 passed

PR Checklist

• PR description clearly explains what and why.
• PR Follows TRT-LLM CODING GUIDELINES to the best of my knowledge.
• Test cases are provided for new code paths.
• Any new dependencies have been scanned for license and vulnerabilities.
• CODEOWNERS updated if ownership changes (N/A).
• Documentation updated as needed (N/A).