[SPARK-50824][PYTHON] Avoid importing optional Python packages for ch…

…ecking

### What changes were proposed in this pull request?

This PR proposes to avoid importing optional Python packages for checking, by using `importlib.util.find_spec` instead of actually loading/importing the package.

### Why are the changes needed?

a409199 changed to import optional dependencies in main code. After that, technically f223b8d broke the Python Spark Core tests, (because now we will import `pyspark.testing`, and it will import optional dependencies) but it did not run the tests.

By importing `deepspeed`, via logger, it can show stdout (https://github.com/microsoft/DeepSpeed/blob/master/accelerator/real_accelerator.py#L182). This broke the test in `pyspark.conf`. After that, the real test failure was found when core change was triggered at 6f3b778. In the PR, build passed because it was before f223b8d was merged.

### Does this PR introduce _any_ user-facing change?

Technically yes. There might be some side effects by importing optional dependencies, and this PR avoid them.

### How was this patch tested?

CI in this PR should verify it.

### Was this patch authored or co-authored using generative AI tooling?

No.

Closes #49500 from HyukjinKwon/SPARK-50824.

Authored-by: Hyukjin Kwon <gurwls223@apache.org>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>

python/pyspark/testing/utils.py

@@ -52,13 +52,9 @@
 
 
 def have_package(name: str) -> bool:
-    try:
-        import importlib
+    import importlib
 
-        importlib.import_module(name)
-        return True
-    except Exception:
-        return False
+    return importlib.util.find_spec(name) is not None
 
 
 have_numpy = have_package("numpy")