Add term processing module and counters

[artvandelay]

Motivation

• Provide a dedicated text-processing layer to tokenize added/removed diff text and surface candidate terms for signals.
• Capture multi-token phrases and candidate proper nouns/capitalized phrases for better term detection and entity signals.
• Maintain time-windowed counters to compare short-term spikes against 7d/30d baselines for anomaly detection.

Description

• Add src/nlbt/terms/tokenizer.py with tokenize, tokenize_text, and extract_capitalized_phrases using a word regex for robust tokenization.
• Add src/nlbt/terms/ngrams.py with generate_ngrams (1–4 grams by default) and extract_term_candidates to produce n-grams and proper-noun candidates.
• Add src/nlbt/terms/diff.py with tokenize_diff and extract_diff_terms helpers to process added/removed text as diffs.
• Add src/nlbt/terms/counters.py implementing TermCounterStore and TermBucketStats for time-bucketed storage (hourly by default) and rollups via get_rollups (24h/7d/30d) while tracking distinct pages and editors.

Testing

• Added tests/test_terms.py which exercises diff tokenization, n-gram extraction, and proper-noun capture via tokenize_diff and extract_diff_terms.
• Ran python tests/test_terms.py and the test script completed successfully (all assertions passed).

---

Codex Task