Fix ptxas compilation on sm103 for triton kernels (#2539)

* add triton ptxas path  for gb300 to find where it is to avoid compilation errors
Signed-off-by: tdophung <tdophung@nvidia.com>

* add these flags in advance to preven future breaks when ops are extended to multi gpus
Signed-off-by: tdophung <tdophung@nvidia.com>

* add this also to L1
Signed-off-by: tdophung <tdophung@nvidia.com>

---------
Signed-off-by: tdophung <tdophung@nvidia.com>

qa/L0_jax_distributed_unittest/test.sh

 # Copyright (c) 2022-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # See LICENSE for license information.
+export TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
 
 function error_exit() {
     echo "Error: $1"

qa/L0_jax_unittest/test.sh

 # Copyright (c) 2022-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # See LICENSE for license information.
+export TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
 
 set -x
 

qa/L1_jax_distributed_unittest/test.sh

 # Copyright (c) 2022-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # See LICENSE for license information.
+export TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
 
 function test_fail() {
     RET=1

qa/L2_jax_distributed_unittest/test.sh

 # Copyright (c) 2022-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # See LICENSE for license information.
+export TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
 
 set -xe
 

qa/L2_jax_unittest/test.sh

 # Copyright (c) 2022-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # See LICENSE for license information.
+export TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
 
 set -x