Remove max_new_tokens and add default tool_call_parser, reasoning_parser, enable_thinking values

restore reasoning: enable_thinking default true + preserve_thinking

spinquantless FP8 (no SpinQuant rotation), 256K context - fixes agentic looping

Enable thinking by default, preserve reasoning; drop max_new_tokens cap

Mark </assistant> (token 24) as special in tokenizer.json

Mark </assistant> (token 24) as special to match internal serving

Fix chat template: preserve reasoning across turns (llama.cpp --reasoning-preserve)

Laguna S 2.1 initial release