training: push adapters to HF Hub after SFT + each GRPO stage

grpo: skip-SFT continuation script + completion-shape fix

train_grpo: import unsloth at module top before trl

sft_warmstart: import unsloth first; batched formatting_func

sft_warmstart: hardcode Qwen2.5 eos_token <|im_end|>

train_grpo: same eos placeholder fix for GRPO

sft_warmstart: replace unsloth's <EOS_TOKEN> placeholder with <|im_end|>

sft_warmstart: pass real eos_token (trl 0.24 sentinel bug)

sft_warmstart: trl 0.24 API (SFTConfig + processing_class)

OpenSOC v1