[libcu++] Correctly handle extended lambda in cuda::launch

[pciolkosz]

There is no way to inspect arguments of an extended lambda in the host code, so we can't automatically dispatch to either pass the kernel config into the lambda or not.
For now the safe option is to require the user to explicitly pass in the config twice, which is not ideal, but lambda kernels are expected to be short and might not need the config object.
Added a test for extended lambda, had to move to lit test to be able to specify the nvcc flag.

This PR also changes some of the common testing headers, because I wanted to use the pinned type in lit test without including the c2h headers.

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 1h 24m: Pass: 100%/91 | Total: 15h 18m | Max: 42m 46s | Hits: 99%/214165

See results here.