extensions/nanoarrow_device/src/nanoarrow/nanoarrow_device_cuda.c (2 lines):
	- line 170: // TODO: All these buffer copiers would benefit from cudaMemcpyAsync but there is
	- line 315: // TODO: We can theoretically also do a move from CUDA_HOST to CUDA


extensions/nanoarrow_device/src/nanoarrow/nanoarrow_device.c (1 line):
	- line 385: // TODO: This is not actually sufficient for CUDA, where the synchronization


r/src/materialize_dbl.h (1 line):
	- line 95: // TODO: implement bounds check for int64 and uint64, but instead


extensions/nanoarrow_device/src/nanoarrow/nanoarrow_device_metal.cc (1 line):
	- line 299: // TODO: sync events for Metal are harder than for CUDA


r/src/convert.c (1 line):
	- line 291: // TODO: Currently we error at the materialize stage if a conversion is not possible;