Make warpspeed scan work in CCCL.C

[bernhardmgruber]

After #6811 is merged, we should expose the new scan implementation to CCCL.C as well. Since this requires us to write and extend all the policy wrappers, we should probably first port cub::DeviceScan to the new tuning API design.

Depends on:

• Implement the new tuning API for DeviceScan #7521

[bernhardmgruber]

I hit a roadblock here when investigating the crash:

./bin/cccl.c.parallel.test.scan
...
========= Out-of-range shared or local address
=========     at void cuda::ptx::__4::mbarrier_init<void>(unsigned long long *, const unsigned int &)+0x770 in mbarrier_init.h:19
=========     by thread (0,0,0) in block (0,0,0)
=========         Device Frame: cub::_V_300300_SM_1200::detail::scan::SyncHandler::clusterInitSync(cub::_V_300300_SM_1200::detail::scan::SpecialRegisters, cub::_V_300300_SM_1200::detail::scan::SkipSync)+0x600 in SyncHandler.h:134

The problem is that the JIT compilation determines that we can use the lookahead scan, while the precompiled dispatching logic does not, since cccl_indirect_iterator_t is not a contiguous_iterator. So the host sets up the old kernel, but then launches the new one, and we have insufficient SMEM (and lots of other bugs).

Before touching all the PTX-JSON stuff to make that work, I think it would be better to port DeviceScan to the new tuning API.