c.parallel: device wrappers as code, not format strings

[griwes]

Description

So far, even with the recent refactoring, the code for the device wrappers lived in format strings. This has a number of downsides, and this PR aims to resolve those.

The wrappers are now actual C++ templates, can be edited with the help of LSP and code formatters, and attempts to obtain their type names will now do some rudimentary type checking in the host code. This is not perfect, as things like cccl_op_t carry basically no type information, but is an improvement over the status quo.

Also added is a CMake target that bundles all the wrappers into a file that contains their contents, preprocessed to a degree, wrapped into a string, so all the code can be simply added into the NVRTC TU wherever needed.

Resolves #2525
Resolves #2665
Resolves #2918

Checklist

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

[github-actions]

🟩 CI finished in 51m 21s: Pass: 100%/3 | Total: 1h 03m | Avg: 21m 13s | Max: 50m 58s

• 🟩 cccl_c_parallel: Pass: 100%/2 | Total: 12m 42s | Avg: 6m 21s | Max: 10m 39s

  🟩 cpu
    🟩 amd64              Pass: 100%/2   | Total: 12m 42s | Avg:  6m 21s | Max: 10m 39s
  🟩 ctk
    🟩 12.6               Pass: 100%/2   | Total: 12m 42s | Avg:  6m 21s | Max: 10m 39s
  🟩 cudacxx
    🟩 nvcc12.6           Pass: 100%/2   | Total: 12m 42s | Avg:  6m 21s | Max: 10m 39s
  🟩 cudacxx_family
    🟩 nvcc               Pass: 100%/2   | Total: 12m 42s | Avg:  6m 21s | Max: 10m 39s
  🟩 cxx
    🟩 GCC13              Pass: 100%/2   | Total: 12m 42s | Avg:  6m 21s | Max: 10m 39s
  🟩 cxx_family
    🟩 GCC                Pass: 100%/2   | Total: 12m 42s | Avg:  6m 21s | Max: 10m 39s
  🟩 gpu
    🟩 v100               Pass: 100%/2   | Total: 12m 42s | Avg:  6m 21s | Max: 10m 39s
  🟩 jobs
    🟩 Build              Pass: 100%/1   | Total:  2m 03s | Avg:  2m 03s | Max:  2m 03s
    🟩 Test               Pass: 100%/1   | Total: 10m 39s | Avg: 10m 39s | Max: 10m 39s
  

• 🟩 python: Pass: 100%/1 | Total: 50m 58s | Avg: 50m 58s | Max: 50m 58s

  🟩 cpu
    🟩 amd64              Pass: 100%/1   | Total: 50m 58s | Avg: 50m 58s | Max: 50m 58s
  🟩 ctk
    🟩 12.6               Pass: 100%/1   | Total: 50m 58s | Avg: 50m 58s | Max: 50m 58s
  🟩 cudacxx
    🟩 nvcc12.6           Pass: 100%/1   | Total: 50m 58s | Avg: 50m 58s | Max: 50m 58s
  🟩 cudacxx_family
    🟩 nvcc               Pass: 100%/1   | Total: 50m 58s | Avg: 50m 58s | Max: 50m 58s
  🟩 cxx
    🟩 GCC13              Pass: 100%/1   | Total: 50m 58s | Avg: 50m 58s | Max: 50m 58s
  🟩 cxx_family
    🟩 GCC                Pass: 100%/1   | Total: 50m 58s | Avg: 50m 58s | Max: 50m 58s
  🟩 gpu
    🟩 v100               Pass: 100%/1   | Total: 50m 58s | Avg: 50m 58s | Max: 50m 58s
  🟩 jobs
    🟩 Test               Pass: 100%/1   | Total: 50m 58s | Avg: 50m 58s | Max: 50m 58s
  

👃 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: 3)

#	Runner
2	linux-amd64-gpu-v100-latest-1
1	linux-amd64-cpu16

[shwina]

I can't provide feedback on the implementation, but it's definitely a quality-of-life improvement to be able to define the kernels in code rather than as strings.