WIP: software/llamacpp: expose llama.cpp sampling and reasoning_budget.
top-k/top-p/temp/min-p is for sampling.
reasoning_budget is for control thinking mode
dry_multiplier is DRY sampling multiplier, can control repeat penalty
top-k/top-p/temp/min-p is for sampling.
reasoning_budget is for control thinking mode
dry_multiplier is DRY sampling multiplier, can control repeat penalty