fix: check wc.status before opcode, use MORI logger in proxy

[itej89]

Summary

Addresses PR ROCm#558 review comments #1, #3, and #7.

• feat: CPU proxy RDMA for EP on rail-isolated clusters (v3-clean) #7: Check wc.status before wc.opcode in DrainCq — on error, opcode may be garbage
• feat(io): fuse RDMA notification into write-with-immediate #1: Null-check atomic target address from SEND_WITH_IMM payload
• Feat/ibgda proxy parallel qp #3: Replace fprintf with MORI_CORE_ERROR in proxy_thread.cpp, MORI_SHMEM_INFO in init.cpp

Test plan

• EP benchmark on DO AINIC: Dispatch Pass, Combine Pass
• Dispatch avg 32.45 GB/s, Combine avg 56.77 GB/s (consistent)

🤖 Generated with Claude Code