refactor: simplify BaseFactory and separate ModelFactory from AutoModel

- Extract _resolve_base_type and _validate_component as module-level helpers
- Replace ForwardRef._evaluate private API with eval in module namespace
- Remove broad except Exception in __init_subclass__, _component_base always set
- Replace direct _entries mutation in strategy.py with register() call form
- Remove dead TOKENIZER_CLASSES registry from AutoTokenizer
- Extract ModelFactory(BaseFactory[nn.Module]) as pure factory
- AutoModel now inherits only nn.Module, no factory state
- Move @AutoModel.register to @ModelFactory.register in transformer.py and encoder.py

tests/module/test_encoder.py

@@ -7,7 +7,7 @@ import safetensors.torch as st
 import torch
 
 from astrai.config.model_config import EncoderConfig
-from astrai.model.automodel import AutoModel
+from astrai.model.automodel import ModelFactory
 from astrai.model.encoder import EmbeddingEncoder
 from tests.helpers import TINY_CONFIG, assert_state_dicts_equal
 
@@ -69,8 +69,8 @@ def test_encoder_normalize(device):
 
 
 def test_encoder_register():
-    assert AutoModel.is_registered("embedding")
-    cls = AutoModel.get_component_class("embedding")
+    assert ModelFactory.is_registered("embedding")
+    cls = ModelFactory.get_component_class("embedding")
     assert cls is EmbeddingEncoder
 
 

tests/trainer/test_online_strategy.py

@@ -92,8 +92,8 @@ def _make_dpo(device, executor=None):
 def test_factory_registers_online_aliases():
     assert StrategyFactory.is_registered("online_grpo")
     assert StrategyFactory.is_registered("online_dpo")
-    assert StrategyFactory._entries["online_grpo"] is GRPOStrategy
-    assert StrategyFactory._entries["online_dpo"] is DPOStrategy
+    assert StrategyFactory.get_component_class("online_grpo") is GRPOStrategy
+    assert StrategyFactory.get_component_class("online_dpo") is DPOStrategy
 
 
 def test_grpo_supports_online(device):