perf: increase llama_cpp initialization and prefill timeouts to 120s for worst-case CPU performance

perf: increase llama_cpp timeout thresholds to 60s for slow CPU architectures

fix: improve mock backend conversation memory and history-aware responses

fix: Update mock ledger parser regex to prevent matching rupees as dates and support dynamic items (e.g. cotton)

docs: Update README with enhancements; implement GGUF latency fallback, KV cache thread-safety lock, mock structured parser, start-from-scratch reset, mobile CSS responsiveness, and nudge parser fixes

fix: resolve prefix repetition loop by yielding deltas only in stream

limit n_threads to 1 on HF Spaces to prevent repack scheduler crashes

mock to realtime local model execution with llama.cpp

mock run working