fix: resolve audited training, import, and serving bugs

- shard the Muon Newton-Schulz orthogonalization over the FSDP mesh instead of partial local slices
- import HF checkpoints faithfully: per-head RoPE permutation for q/k projections and qk-norm, qwen3, shared experts, and qk-norm before RoPE (changes numerics for existing use_qk_norm checkpoints)
- make preprocessing and resume self-contained: backfill realigned bucket keys by semantics (masks ones, rest zeros) and snapshot tokenizer files into every checkpoint
- keep RL consistent: sync the offline GRPO old_model each optimizer step and validate online strategies through a public one-off-rollout hook that leaves the replay cache untouched
- fix streaming serving: withhold partial tool-call prefixes with a stream-end flush, stream tool-call arguments from the raw source span, and terminate SSE frames with a blank line
- fix sampling semantics: capture logprobs before top-k/top-p mutate logits in place and detect greedy pipelines polymorphically instead of isinstance bookkeeping

astrai/inference/network/protocol.py

@@ -20,8 +20,10 @@ def sse_event(data: Dict[str, Any], event: Optional[str] = None) -> str:
     if event:
         lines.append(f"event: {event}")
     lines.append(f"data: {json.dumps(data, ensure_ascii=False)}")
-    lines.append("")
-    return "\n".join(lines)
+    # The SSE spec dispatches an event only at a blank line, so the frame
+    # must end with "\n\n" (a single trailing newline keeps clients waiting
+    # and concatenates consecutive events into one corrupt payload).
+    return "\n".join(lines) + "\n\n"
 
 
 def sse_done() -> str: