Qwen 3.5 MoE and 3.6 MoE FP8 support restored in v5.6.2 — kernel config/read fixes

AI Impact Summary

Patch v5.6.2 fixes FP8 breakage for Qwen 3.5 MoE and Qwen 3.6 MoE (text-only) by correcting kernel configuration parsing and error handling (referenced in #45610). Upgrading to this release restores FP8 workflows without code changes needed for existing FP8 integrations, though operators should re-run FP8 tests to confirm stability. The change tightens the configuration path and error reporting to reduce the risk of future FP8-related failures in these models.