Is there a hard blocker preventing Muon to be used with gated deltanet and gated attention?

Hi @FDecaYed I would like to consult you on using Muon with linear attention and gated attention.

In https://github.com/NVIDIA/Megatron-LM/commit/5880674a396ed3489d185afaefb1773a6481fe18#diff-42fa19ec8893eabf951a5bb21edfc0dfe7c9c8949d5087ab133f5897ea0e3213 it is mentioned in the arguments line 1203 that
```
assert args.linear_attention_type is None, "Muon optimizer does not support linear attention type for now."
assert not args.attention_output_gate, "Muon optimizer does not support attention output gate for now."
```

However, is there a hard incompatibility, or is it just not supported yet?

Could they simply be routed to Adam while the rest of the layers continue to use Muon?