Add unit tests for the wint2 quantized MoE expert FFN op.
Source: look for moe_expert_ffn_wint2 in custom_ops/gpu_ops/
Registration: custom_ops/gpu_ops/cpp_extensions.cc
Test file: tests/operators/test_moe_expert_ffn_wint2.py
This is a fused MoE FFN with wint2 quantized weights. Reference test: test_fused_moe.py. Cover different expert counts and hidden sizes.
Branch: task/039-moe-expert-ffn-wint2-test