[Python] Fix accidental dependency on pandas

[Tishj]

This PR fixes #5577

I don't believe we run tests on a environment without pandas installed, as it is part of the requirements-dev.txt
So I'm not sure if I can add a test for this, but "it works on my machine" (created a clean venv, verified pandas was not installed, ran the example provided in the issue, and it no longer crashed)

Whereas on master it produced:

(venv) ➜  duckdb git:(master) ✗ python3 tmp/implicit_pandas_dep.py     
Traceback (most recent call last):
  File "/Users/thijs/DuckDBLabs/duckdb/tmp/implicit_pandas_dep.py", line 5, in <module>
    conn.execute("INSERT INTO foo VALUES (?)", (datetime.datetime.now(),))
ModuleNotFoundError: No module named 'pandas'

[Mytherin]

Thanks for the PR! Looks great. One small comment:

[Mytherin]

Should we check if pandas is loaded instead? Or does that not matter because we always call LoadSubtypes?

[Tishj]

Hmm no, LoadSubtypes would not be called in this case because pandas failed to load, but the constructor of PythonImportCacheItem sets object to nullptr, which is what IsLoaded checks for

So it makes no difference if I check IsLoaded on isnull or pandas, if you prefer doing it on pandas instead that's also fine

Re-reading the python import cache code, I do notice that we don't have error handling for the case where a module is loaded, but the attribute it's trying to find does not exist in the module.
Maybe I need to add a try-catch block around source_object.attr(...) in LoadAttribute