feat: remove cuda and add miette error handling

[lucarlig]

Refactor: Migrate from GPU to NPU Architecture and Clean Up Examples

• Renamed all GPU terminology to NPU across the entire codebase (device types, memory spaces, function names, execution contexts)
• Updated device types: gpu_device → npu_device, gpu_alloc_copy → npu_alloc_copy
• Migrated memory spaces: gpu.global → npu.global, gpu.shared → npu.shared, gpu.grid → npu.grid
• Updated function names: gpu_* → npu_* for all kernel functions
• Refactored execution contexts: gpu.block → npu.block, gpu.grid → npu.grid
• Removed 30+ outdated CUDA examples from cuda-examples/asplos/ directory
• Deleted benchmark result files (cuda-examples/res/) containing generated performance metrics
• Disabled problematic examples by renaming .desc files to .desc.off
• Added new NPU-focused examples demonstrating vector operations (vadd, vmul, vdiv)
• Updated remaining examples to use NPU terminology and constructs
• Removed .cargo/config.toml and restored compiler warnings (unused_variables, dead_code, unreachable_code)
• Deleted LICENSE file from repository
• Cleaned up trailing whitespace in example files
• Added new error examples (invalid_type_add.desc, missing_main.desc) for better testing coverage
• Updated existing error examples to use NPU terminology
• Aligned codebase with new target architecture (Huawei Ascend NPUs via MLIR/AscendNPU-IR)
• Removed legacy CUDA-specific code and improved maintainability