Refactor model references and enhance ZeroGPU support; update README and requirements for HF

[Improvement - alpha] Implement RSARerankingCached for improved efficiency in interactive review processing; update import in interactive_processor.py to use cached version.

Enhance interactive review processing by adding thread management for polarity and topic predictions; set optimal thread count for SLURM environments; update loss function to ignore padding tokens.

Reduce max_length for model inputs to 256 tokens for improved performance; add BetterTransformer support for CPU speedup in interactive review processing.

Add progress tracking for RSA computation in background and interactive processor

Add informativeness computation and enhance RSA processing: implement sentence ranking by informativeness and uniqueness, and improve HTML rendering for agreement and divergence summaries.

Add centralized sentence filtering for RSA and polarity scoring; enhance interactive review processor to filter noise sentences and improve summary card formatting

Update model loading to prioritize local trained models for improved accuracy; adjust configuration for model paths and fallback options

Add device-aware RSA optimizations for CPU/GPU

Fix model loading utility and improve error handling in scoring pipeline

Add scoring utilities and unified scoring pipeline for ICLR review data

Remove dependencies on glimpse repo for interface.