Update range of gpu arch

[yf711]

Description

Remove deprecated gpu arch and reduce nuget/python package size (latest TRT supports sm75 Turing and newer arch)

Test on pkg CI	Python-cuda12	Nuget-cuda12
Before	Linux: 279MB Win: 267MB	Linux: 247MB Win: 235MB
After	Linux: 174MB Win: 162MB	Linux: 168MB Win: 156MB

Motivation and Context

[tianleiwu]

If we drop older arch, shall we also drop ort package for cuda 11.8 in next release?

[snnn]

If we drop older arch, shall we also drop ort package for cuda 11.8 in next release?

I highly recommend doing so. Now we only have two people working on build pipelines. We should focus more on the main targets.

[snnn]

/azp run Win_TRT_Minimal_CUDA_Test_CI

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[yf711]

After testing, adding sm90 to build arch list is causing issues to cuda 11.8+cudnn8 alt pkg build on windows,
which is likely because cudnn8 is deprecated by blackwell. cuda 12 pkg build is not affected.

To support sm90, we can choose to support cuda12 only, or we might need to update current cuda 11.8 env with cudnn9

[snnn]

CUDA 11.8 with cudnn9 doesn't work. I tried.

I hit the following compilation error when compiling cudnn_flash_attention.cu

/build/Release/_deps/cudnn_frontend-src/include/cudnn_frontend/graph_interface.h:519:27:   required from here
/build/Release/_deps/cudnn_frontend-src/include/cudnn_frontend/thirdparty/nlohmann/json.hpp:9132:68: error: static assertion failed: Missing/invalid function: bool boolean(bool)
 9132 |     static_assert(is_detected_exact<bool, boolean_function_t, SAX>::value,

Therefore, I suggest giving up on that.