♻️ refactor: create ops module and move chunked_attention

- Create nanovllm/ops/ module for low-level attention operators
- Move chunked_attention.py from kvcache/ to ops/
- Update imports in full_policy.py (3 locations)
- Fix: remove dead code in OffloadEngine.reset() referencing
  non-existent layer_k/v_buffer_a/b attributes

Verified with needle test (32K offload): PASSED

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

nanovllm/ops/__init__.py

@@ -0,0 +1,19 @@
+"""
+Operators module for nano-vLLM.
+
+This module contains low-level attention operators and kernels.
+"""
+
+from nanovllm.ops.chunked_attention import (
+    flash_attn_with_lse,
+    merge_attention_outputs,
+    chunked_attention_varlen,
+    ChunkedPrefillState,
+)
+
+__all__ = [
+    "flash_attn_with_lse",
+    "merge_attention_outputs",
+    "chunked_attention_varlen",
+    "ChunkedPrefillState",
+]