feat: add optional TextBlob and HuggingFace backends to classify_sentiment

[agentksimha]

• Added use_textblob and use_huggingface boolean flags to classify_sentiment() — both default to False, keeping VADER as the default with no behaviour change
• hf_pipeline and textblob_analyzer initialised as None at module level and globalised inside the function; each loads only once on first use, avoiding redundant re-instantiation across .apply() calls
• Imports for transformers and textblob are lazy — pipeline won't break if neither package is installed