coverage: add cuda.core tests for graph, IPC, launcher, program, and DLPack

[rluo8]

Description

Adds focused unit tests across cuda.core, targeting input-validation branches, error paths, and object-lifecycle behavior that were previously uncovered. All tests are real-path: host-only tests exercise pure helpers / argument validation directly, and GPU-required tests use the existing fixtures and skips. No production changes.

Measured effect on cuda/core/ line coverage: about +1.65%, +208 lines. Largest gains in _program.pyx, graph/_subclasses.pyx, _memory/_ipc.pyx, _memoryview.pyx, _kernel_arg_handler.pyx, and graph/_graph_builder.pyx.
Note that the coverage numbers are for cuda/core/ only, combined across local systems with Linux, Windows, H800, B10, and B100.

File	Tests
graph/test_graph_builder.py	join type check; __cuda_stream__ after close; capture guards (end_building / debug_dot_print / callback / create_condition / embed / if_then); factory-only GraphBuilder/Graph; close() ends an active capture
graph/test_graph_definition.py	HOST AllocNode reconstructs memory_type from the driver via a pred walk (CUDA_ERROR_NOT_SUPPORTED skip)
graph/test_graph_definition_lifetime.py	pred-chain _create_from_driver for memcpy / event record / event wait / free
graph/test_graph_node_update.py	memset/memcpy owner + no-op + 2D geometry updates; factory-only ExecutableGraphNode; executable-node repr; clustered/cooperative update reject; ctypes host-callback repr
memory_ipc/test_errors.py	factory-only IPC types; raw-fd from_allocation_handle; ForkingPickler fd dup; registry dedup of repeated imports (Linux)
test_launcher.py	numpy scalar-subclass prepare_numpy_arg fallback (kernel readback); CUgraphConditionalHandle packed as uint64
test_memory.py	Buffer() forbidden; context-manager close; abstract MemoryResource stubs; copy_to() without a memory resource
test_optional_dependency_imports.py	NVVM import rejects cuda-bindings < 12.9 and caches the failure
test_program.py	nested define_macro error; each NVVM-unsupported option; NVRTC as_bytes sequence flags; debug temp-write fallback; use_libdevice resolves __nv_sin
test_utils.py	CAI-only proxy .view(); view(dtype=) itemsize mismatch
test_utils_dlpack.py	CPU array-interface export; Buffer re-view owner reuse; CUDA stream=None; versioned readonly; malformed dtype; C-exchange API NULL/scalar/device-accessible cases