Introduce cuda.cooperative overloads not requiring temporary storage

[gevtushenko]

Description

closes #2527

This PR introduces versions of cooperative algorithms that do not require temporary storage. This is a quick fix for temporary storage alignment issues when having more than one shared memory array.

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[github-actions]

🟩 CI finished in 14m 16s: Pass: 100%/1 | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s

• 🟩 pycuda: Pass: 100%/1 | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s

  🟩 cpu
    🟩 amd64              Pass: 100%/1   | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s
  🟩 ctk
    🟩 12.5               Pass: 100%/1   | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s
  🟩 cudacxx
    🟩 nvcc12.5           Pass: 100%/1   | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s
  🟩 cudacxx_family
    🟩 nvcc               Pass: 100%/1   | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s
  🟩 cxx
    🟩 GCC13              Pass: 100%/1   | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s
  🟩 cxx_family
    🟩 GCC                Pass: 100%/1   | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s
  🟩 gpu
    🟩 v100               Pass: 100%/1   | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s
  🟩 jobs
    🟩 Test               Pass: 100%/1   | Total: 14m 16s | Avg: 14m 16s | Max: 14m 16s
  

👃 Inspect Changes

Modifications in project?

	Project
	CCCL Infrastructure
	libcu++
	CUB
	Thrust
	CUDA Experimental
+/-	pycuda
	CCCL C Parallel Library

Modifications in project or dependencies?

	Project
	CCCL Infrastructure
	libcu++
	CUB
	Thrust
	CUDA Experimental
+/-	pycuda
	CCCL C Parallel Library

🏃‍ Runner counts (total jobs: 1)

#	Runner
1	linux-amd64-gpu-v100-latest-1

[gevtushenko]

A few things to consider before merging:

a) sync is required before subsequent calls, which is not obvious, so we might need to add a sync inside the call
b) warp scope would allocate temp storage for a single warp only, so we either should over allocate temporary storage for 1024 / warp size warps, or disallow temporary storage-free API at warp scope

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[gevtushenko]

/ok to test

[github-actions]

🟩 CI finished in 23m 47s: Pass: 100%/1 | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s

• 🟩 python: Pass: 100%/1 | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s

  🟩 cpu
    🟩 amd64              Pass: 100%/1   | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s
  🟩 ctk
    🟩 12.6               Pass: 100%/1   | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s
  🟩 cudacxx
    🟩 nvcc12.6           Pass: 100%/1   | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s
  🟩 cudacxx_family
    🟩 nvcc               Pass: 100%/1   | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s
  🟩 cxx
    🟩 GCC13              Pass: 100%/1   | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s
  🟩 cxx_family
    🟩 GCC                Pass: 100%/1   | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s
  🟩 gpu
    🟩 v100               Pass: 100%/1   | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s
  🟩 jobs
    🟩 Test               Pass: 100%/1   | Total: 23m 47s | Avg: 23m 47s | Max: 23m 47s
  

👃 Inspect Changes

Modifications in project?

	Project
	CCCL Infrastructure
	libcu++
	CUB
	Thrust
	CUDA Experimental
+/-	python
	CCCL C Parallel Library
	Catch2Helper

Modifications in project or dependencies?

	Project
	CCCL Infrastructure
	libcu++
	CUB
	Thrust
	CUDA Experimental
+/-	python
	CCCL C Parallel Library
	Catch2Helper

🏃‍ Runner counts (total jobs: 1)

#	Runner
1	linux-amd64-gpu-v100-latest-1