Migrate CI configs to CMake presets.

[alliepiper]

Description

Closes #252

MIgrate Thrust and CUB's Github CI config/build/test/ steps to use CMake presets.

[jrhemstad]

An overarching goal is to push as much logic into the cmake presets or cmake targets as possible. In other words, there should be very little logic in the build scripts themselves because that needs to be replicated across both Linux and Windows build scripts. Ideally, cmake should be the source of truth for everything.

• This currently breaks how clangd is configured because it expects the compiler commands database to be in build/latest. This is done via a symlink in the build_common.sh. Instead, the symlinking should be done as part of the cmake configure. We should change this symlinking to instead symlink the compile-commands.json to the cccl source dir so that it can work outside of VSCode/devcontainer too.
• The sccache stats logic from build_common.sh should be made part of the cmake build such that it will always prints sccache stats when you do a build via one of the presets. (Deferred to future work)
• It would be very nice if we can make sure the cmake presets work outside of the context of a devcontainer if possible, even if at some reduced functionality (e.g., no sccache).
• Verify that CTEST_PARALLEL_LEVEL doesn't actually make the tests slower
• To maximize the effectiveness of sccache, we want CI and local devs to build with the same set of architectures whenever possible. Therefore, those should be set in the cmake presets to be the same between local dev/CI.
• The build directory for the cmake presets should ideally check if running in a devcontainer and set to a unique build directory to avoid conflicts when using multiple containers.

[jrhemstad]

re: architecture, I remembered that for the CI jobs, that is currently set in the matrix file here:
https://github.com/NVIDIA/cccl/blob/486e55792b4f516a16c58c3d303b7a48e724471e/ci/matrix.yaml#L30C128-L30C134

I want to have a single source of truth whenever possible, so I think we could eliminate the gpu_build_archs from the matrix file and just defer to whatever is set in the preset.

Alternatively, the cmake presets could parse the architecture from the matrix.yaml file (assuming we had a top-level entry for the gpu archs).

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[jrhemstad]

Can we also update the contributing guide section that describes how to use the scripts: https://github.com/NVIDIA/cccl/blob/main/CONTRIBUTING.md#building-and-testing

[jrhemstad]

question (for self): Does this overwrite the CXX envvar after running the script?

Addressed requested changes

[alliepiper]

This is ready for another round of reviews.