reuse T4 on Linux GPU (#19879)

### Description

### Motivation and Context
Linux GPU test on A10 isn't very stable

Author: mszhanyi

tools/ci_build/github/azure-pipelines/linux-gpu-ci-pipeline.yml

@@ -150,7 +150,7 @@ stages:
               --enable_cuda_profiling --enable_cuda_nhwc_ops \
               --enable_pybind --build_java \
               --use_cache \
-              --cmake_extra_defines  CMAKE_CUDA_ARCHITECTURES=86; \
+              --cmake_extra_defines  CMAKE_CUDA_ARCHITECTURES=75; \
                 ccache -sv; \
                 ccache -z"
       workingDirectory: $(Build.SourcesDirectory)
@@ -183,7 +183,7 @@ stages:
       skipComponentGovernanceDetection: true
     workspace:
       clean: all
-    pool: onnxruntime-Linux-GPU-A10
+    pool: onnxruntime-Linux-GPU-T4
     steps:
     - checkout: self
       clean: true