fix(handlers): EMAHandler warmup now respects custom name parameter

When attach() is called with a custom name (e.g. for GAN generators),
the momentum warmup scheduler was hardcoded to write to
engine.state.ema_momentum instead of engine.state.{name}. This meant
warmup had no effect with custom names.

Fix: pass the name parameter through to LambdaStateScheduler.

Author: VirajMishra1

ignite/handlers/ema_handler.py

@@ -248,7 +248,7 @@ def attach(
             setattr(engine.state, name, self.momentum)
 
         if self._momentum_lambda_obj is not None:
-            self.momentum_scheduler = LambdaStateScheduler(self._momentum_lambda_obj, param_name="ema_momentum")
+            self.momentum_scheduler = LambdaStateScheduler(self._momentum_lambda_obj, param_name=name)
 
             # first update the momentum and then update the EMA model
             self.momentum_scheduler.attach(engine, event)