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/inference/runtime/__init__.py

@@ -0,0 +1,25 @@
+"""Execution primitives: forward passes, CUDA graphs, and sampling."""
+
+from astrai.inference.runtime.executor import Executor
+from astrai.inference.runtime.graph import CudaGraphContext
+from astrai.inference.runtime.sample import (
+    BaseSamplingStrategy,
+    FrequencyPenaltyStrategy,
+    SamplingPipeline,
+    TemperatureStrategy,
+    TopKStrategy,
+    TopPStrategy,
+    sample,
+)
+
+__all__ = [
+    "Executor",
+    "CudaGraphContext",
+    "BaseSamplingStrategy",
+    "FrequencyPenaltyStrategy",
+    "SamplingPipeline",
+    "TemperatureStrategy",
+    "TopKStrategy",
+    "TopPStrategy",
+    "sample",
+]