GPU: Check runtime calls and kernel launches in release builds

Route Tensor GPU runtime calls and kernel launches through a configurable error check. The default reports and terminates on failure even with NDEBUG; retain opt-out and user override hooks. Test successful and invalid launches, argument evaluation, and release-mode overrides.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

16 files changed