refactor: split infer core into subpackages by concern

- Eliminate core/ directory into cache/, runtime/, network/ subpackages plus flat modules
- Split cache.py (647 lines) into cache/{buffer,strategy,pool}.py by layer
- Add explicit ContiguousStrategy, make AllocationStrategy a real ABC
- Move TaskCacheState to cache/strategy.py, drop string forward references
- Rename api/ to network/, server.py to app.py
- Move sample.py into runtime/ alongside executor and graph
- Simplify TaskCacheManager.__init__ to single pool param
- Expose pool.strategy and pool.req_pool as public properties
- Fix KVCache import in attention_backend.py (TYPE_CHECKING guard)
- Fix steady-state decode reading uninitialized position_ids on first step

astrai/extension/attention_backend.py

@@ -51,8 +51,11 @@ from astrai.extension.loader import is_available
 from astrai.factory import BaseFactory
 
 if TYPE_CHECKING:
-    from astrai.inference.core.cache import KVCache
+    from astrai.inference.cache import KVCache
 
+
+_default_backend: Optional["AttentionBackend"] = None
+_default_backend_lock = threading.Lock()
 _current_backend: contextvars.ContextVar["AttentionBackend"] = contextvars.ContextVar(
     "attn_backend"
 )
@@ -101,10 +104,6 @@ class ATTN_BACKEND(enum.Enum):
     FLASH = "flash"
 
 
-_default_backend: Optional["AttentionBackend"] = None
-_default_backend_lock = threading.Lock()
-
-
 def _priority_backends() -> list["AttentionBackend"]:
     """Available backends in priority order: cuda -> flash -> torch."""
     backends: list[AttentionBackend] = []