fix(core.utils): lazy-import sqlite3 to avoid libsqlite3 dep at import

sqlite3 was imported at module level, which fails on CI containers
that don't have libsqlite3.so.0 installed (e.g., CUDA 12.9.1 test
images). Move the import into SQLiteProgramCache.__init__ so the
module is loadable even without sqlite3 — only constructing a
SQLiteProgramCache requires it.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: cpcloud

cuda_core/cuda/core/utils/_program_cache.py

@@ -21,7 +21,6 @@
 import hashlib
 import os
 import pickle
-import sqlite3
 import tempfile
 import time
 from pathlib import Path
@@ -336,6 +335,9 @@ def __init__(
         self._path = Path(path)
         self._path.parent.mkdir(parents=True, exist_ok=True)
         self._max_size_bytes = max_size_bytes
+        import sqlite3
+
+        self._sqlite3 = sqlite3
         self._conn: Optional[sqlite3.Connection] = None
         self._open()
 
@@ -346,7 +348,7 @@ def _open(self) -> None:
         # each statement is its own transaction; ``check_same_thread=False``
         # lets a cache be created in one thread and used from another (writes
         # are still serialised by sqlite's own lock).
-        self._conn = sqlite3.connect(
+        self._conn = self._sqlite3.connect(
             self._path,
             isolation_level=None,
             check_same_thread=False,
@@ -385,7 +387,7 @@ def close(self) -> None:
             finally:
                 self._conn = None
 
-    def _require_open(self) -> sqlite3.Connection:
+    def _require_open(self):
         if self._conn is None:
             raise RuntimeError("SQLiteProgramCache is closed")
         return self._conn