feat(ltx2): use default VAEShardingSpecs for all strategies

Author: Perseus14

src/maxdiffusion/models/ltx2/logical_sharding_ltx2.py

@@ -106,7 +106,7 @@ class VAEShardingSpecs:
             out_kernel=("heads", None),
             out_bias=(None,),
         ),
-        "vae": VAEShardingSpecs(vae_conv_kernel=("batch", None, None, None)),
+        "vae": VAEShardingSpecs(),
     },
     "trillium": {
         "ltx2_dit": LTX2DiTShardingSpecs(
@@ -119,7 +119,7 @@ class VAEShardingSpecs:
             text_encoder_kernel=(None, "embed"),
         ),
         "text_connector": TextConnectorShardingSpecs(),
-        "vae": VAEShardingSpecs(vae_conv_kernel=(None, None, None, None)),
+        "vae": VAEShardingSpecs(),
     },
 }