fix: `nltk` data download path to prevent redundant nested directories (#3546)
Closes #3543.
### Summary
This PR addresses an issue with the NLTK data download process.
Previously, when downloading NLTK data, a nested "nltk_data" directory
was created within the parent "nltk_data" directory if the parent
directory already existed. This redundant directory structure led to two
significant problems:
- errors in checking if data had already been downloaded, potentially
causing redundant downloads in subsequent calls.
- failures in loading models from the downloaded NLTK data due to
incorrect path resolution.
This fix modifies the NLTK data download logic to prevent creation of
unnecessary nested directories. If the download path ends with
"nltk_data" and that directory already exists, we now use the existing
directory instead of creating a new nested one.
### Testing
CI should pass.