#4887 - Conditionally gate SM 100/120 on CUDA 12.8+ in default CMAKE_CUDA_ARCHITECTURES

[Deeven-Seru]

Description

Fixes #4887.

In the TensorRT 11.4 OSS release (#4872), CMAKE_CUDA_ARCHITECTURES was defaulted to include SM 100 and SM 120 unconditionally before CUDAToolkit discovery:

if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
  set(CMAKE_CUDA_ARCHITECTURES 75 80 86 89 90 100 120)
  set(_TRT_CUDA_ARCHS_DEFAULTED TRUE)
endif()

While SM 110 (Thor) was conditionally appended only when CUDAToolkit_VERSION VERSION_GREATER_EQUAL "13.0", SM 100 and SM 120 (Blackwell) require CUDA 12.8+ to be supported by nvcc. When compiling on CUDA toolkits prior to 12.8 (e.g. CUDA 12.0 - 12.6), nvcc fatal errors with:

nvcc fatal : Value 'sm_100' is not defined for option 'gpu-architecture'

TensorRT's internal plugin headers already reflect this requirement via #if CUDA_VERSION >= 12080 in fused_multihead_attention_v2.h.

This PR:

1. Defaults CMAKE_CUDA_ARCHITECTURES to 75 80 86 89 90 prior to toolkit discovery.
2. Conditionally appends SM 100 and SM 120 once CUDAToolkit_VERSION VERSION_GREATER_EQUAL "12.8" is confirmed, matching the existing Thor (110) conditional logic.
3. Fixes the NVRTC version check in samples/python/plugin_utils.py to check (nvrtc_major, nvrtc_minor) >= (11, 1), preventing CUDA 12.0 and 13.0 (nvrtc_minor == 0) from incorrectly disabling CUBIN output.