refactor(muon): remove unused individually-compiled NS functions

torch.compile on muon_update already traces into the NS functions
it calls, so the separately compiled wrappers are redundant.

Signed-off-by: Ma, Guokai <guokai.ma@gmail.com>

Author: delock

deepspeed/runtime/zero/muon/original_muon.py

@@ -151,9 +151,6 @@ def _muon_update(grad, momentum, beta=0.95, ns_steps=5, nesterov=True, ns_method
     return update
 
 
-# Compiled (torch.compile) versions for GPU paths
-zeropower_via_newtonschulz5 = compiler.compile()(_zeropower_via_newtonschulz5)
-zeropower_via_gram_newtonschulz = compiler.compile()(_zeropower_via_gram_newtonschulz)
 muon_update = compiler.compile()(_muon_update)