[hipDNN] Fix test failures in non-GPU environments by handling hipErrorNoDevice (#4426)

Tests were failing with error code 100 ("no ROCm-capable device is
detected") when run in environments without GPUs. This affected all
tests, not just GPU-specific ones, making it impossible to run the test
suite in non-GPU environments.

In non-GPU environments:
1. Any HIP API call triggers HIP runtime initialization
2. HIP detects no GPU and sets `hipErrorNoDevice` (error 100)
3. This error persists and cannot be cleared, even after calling
`hipGetLastError()`
4. The `HipErrorHandler` test listener checks for HIP errors after each
test
5. It detected this persistent error and failed all tests

Modified `HipErrorHandler::OnTestEnd()` to silently accept
`hipErrorNoDevice`:

- Check if the error is `hipErrorNoDevice` after calling
`hipGetLastError()`
- If true, return early without failing the test
- This allows tests to run in non-GPU environments

Author: averinevg

projects/hipdnn/test_sdk/include/hipdnn_test_sdk/utilities/HipErrorHandler.hpp

@@ -25,6 +25,14 @@ class HipErrorHandler : public testing::EmptyTestEventListener
         // Check and clear standard HIP error state
         auto hipError = hipGetLastError();
 
+        // Special case: hipErrorNoDevice cannot be cleared and persists even after
+        // hipGetLastError() is called. In non-GPU environments, we should not fail
+        // the test for this specific error.
+        if(hipError == hipErrorNoDevice)
+        {
+            return;
+        }
+
         // Check and clear extended HIP error state
         auto hipExtError = hipExtGetLastError();