fix: handle punkt_tab fallback in split_sentences for NLTK >= 3.9

[danishashko]

Fixes #1017

NLTK 3.9 moved the default punkt tokenizer from tokenizers/punkt/english.pickle to tokenizers/punkt_tab/english. On a fresh install with NLTK >= 3.9 only punkt_tab is downloaded, so calling split_sentences() raises a LookupError even when NLTK is properly installed.

What this changes:

1. newspaper/nlp.py - split_sentences() now tries the old path first and falls back to punkt_tab if not found. Works with both old and new NLTK versions.
2. download_corpora.py - adds punkt_tab to REQUIRED_CORPORA so it gets downloaded alongside punkt.

PR #1006 already adds punkt_tab to the download list, but the runtime crash in nlp.py still happens if only punkt_tab is available. This fixes both.