Add automatic GPU NCU plan discovery

[yoshifuminakamura]

Summary

• Add NSYS-based GPU kernel discovery and automatic NCU plan generation helpers.
• Connect GENESIS GPU profiling to the generated NCU plan and preserve NSYS discovery metadata in GPU estimation outputs.
• Make result and estimate send jobs work with non-root curl-jq containers by writing generated payloads outside read-only artifact directories.
• Normalize artifact permissions before upload so downstream jobs can read results across runner/container users.

Validation

• WSL bash -n / shellcheck -S error on touched shell scripts.
• GPU estimation shell tests for NCU plan generation and GPU kernel packages.
• Manual MiyabiG CI confirmed run, estimate, and send_estimate paths complete with the generated artifacts.