Provide Run-Length Decode API

With the recent addition of batched copy, we should be able to provide RLD API. This API is needed for tuning infrastructure but should be useful elsewhere. Here's a simplified example from batched copy test to illustrate the idea:

```cpp
const int num_segments = 4;
const int num_items = 12;
thrust::device_vector segment_offsets = {0, 2, 8, num_items};
thrust::device_vector out(num_items);
int *d_out = thrust::raw_pointer_cast(out.data());
int *d_offsets = thrust::raw_pointer_cast(segment_offsets.data());

thrust::counting_iterator iota(0);
auto d_range_srcs = //
thrust::make_transform_iterator(iota, [](int i) {
return thrust::constant_iterator(i);
});
auto d_range_dsts = //
thrust::make_transform_iterator(d_offsets, [d_out](int offset) {
return d_out + offset;
});
auto d_range_sizes = //
thrust::make_transform_iterator(iota, [d_offsets](int i) {
return d_offsets[i + 1] - d_offsets[i];
});

std::uint8_t *d_temp_storage = nullptr;
std::size_t temp_storage_bytes = 0;
cub::DeviceCopy::Batched(d_temp_storage,
temp_storage_bytes,
d_range_srcs,
d_range_dsts,
d_range_sizes,
num_segments);
```

We'll need two overloads:

### Overloads
- [ ] Implement segment-size-based RLD
- [ ] Implement segment-offset-based RLD