refactor: Add ProviderContext for a flexible storage directory

- Introduce ProviderContext class to decouple provider storage paths from absolute paths
- Add storage_dir attribute to StackRunConfig to accept CLI options
- Implement storage directory resolution with prioritized fallbacks:
  1. CLI option (--state-directory)
  2. Environment variable (LLAMA_STACK_STATE_DIR)
  3. Default distribution directory
- Standardize provider signatures to follow context, config, deps pattern
- Update provider implementations to use the new context-based approach
- Add comprehensive tests to verify state directory resolution

llama_stack/providers/inline/inference/meta_reference/__init__.py

@@ -6,15 +6,14 @@
 
 from typing import Any
 
+from llama_stack.providers.datatypes import ProviderContext
+
 from .config import MetaReferenceInferenceConfig
 
 
-async def get_provider_impl(
-    config: MetaReferenceInferenceConfig,
-    _deps: dict[str, Any],
-):
+async def get_provider_impl(context: ProviderContext, config: MetaReferenceInferenceConfig, _deps: dict[str, Any]):
     from .inference import MetaReferenceInferenceImpl
 
-    impl = MetaReferenceInferenceImpl(config)
+    impl = MetaReferenceInferenceImpl(context, config)
     await impl.initialize()
     return impl

llama_stack/providers/inline/inference/meta_reference/inference.py

@@ -50,7 +50,7 @@ from llama_stack.models.llama.llama4.chat_format import ChatFormat as Llama4Chat
 from llama_stack.models.llama.llama4.tokenizer import Tokenizer as Llama4Tokenizer
 from llama_stack.models.llama.sku_list import resolve_model
 from llama_stack.models.llama.sku_types import ModelFamily
-from llama_stack.providers.datatypes import ModelsProtocolPrivate
+from llama_stack.providers.datatypes import ModelsProtocolPrivate, ProviderContext
 from llama_stack.providers.utils.inference.embedding_mixin import (
     SentenceTransformerEmbeddingMixin,
 )
@@ -89,7 +89,8 @@ class MetaReferenceInferenceImpl(
     Inference,
     ModelsProtocolPrivate,
 ):
-    def __init__(self, config: MetaReferenceInferenceConfig) -> None:
+    def __init__(self, context: ProviderContext, config: MetaReferenceInferenceConfig) -> None:
+        self.context = context
         self.config = config
         self.model_id = None
         self.llama_model = None