Add Tesla M10 stress test support

[sdcosmo]

What changed

• add a Tesla M10 profile targeting Maxwell compute capability 5.2
• replace unsupported Maxwell matrix initialization with zero-filled operands
• exercise FP32, complex FP32, FP64, complex FP64, and alternate FP32 layouts
• fix watchdog completion, termination, and multi-GPU index handling
• discover CUDA paths automatically and default the build to sm_52
• document the tested Ubuntu/CUDA/GCC configuration and build commands

Why

The upstream generic profile starts with operations that are unsupported on the Tesla M10. Its custom matrix-fill kernel also fails at runtime with no kernel image is available for execution on the device. Successful tests did not signal watchdog completion, which eventually caused false hang reports during multi-GPU runs.

Validation

• clean CMake build on Ubuntu 24.04 with CUDA 12.0 and GCC/G++ 12
• full -T=1 profile completed on four Tesla M10 8 GiB GPUs
• principal FP32 tests allocate approximately 6.75 GiB per GPU
• validated FP32, C32, FP64, C64, and FP32-NT workloads
• watchdog exits cleanly after the final test

This is an unofficial compatibility fork of NVIDIA/GPUStressTest and retains the upstream MIT and OSS license files.