[WIP] Prevent perfetto-connector from breaking Kokkos Tools CI

[vlkale]

Fix problem with chunk size that occurs with different compile time configurations. This is expected to fix Kokkos Tools Issue #259.

This PR is part of an effort to breaks down PR #262 into two separate pieces. This piece focuses on the Perfetto connector fix, while the other is on allowing for Debian interactive mode and ensuring the latest versions of CMake.

[masterleinad]

Fix problem with chunk size that occurs with different compile time configurations. This is expected to fix Kokkos Tools Issue #259.

Which compiler configuration deosn't allow using sizeof in a constexpr context?

[vlkale]

Fix problem with chunk size that occurs with different compile time configurations. This is expected to fix Kokkos Tools Issue #259.

Which compiler configuration deosn't allow using sizeof in a constexpr context?

@masterleinad It's cuda-11.0.3 from ubuntu, and there could be others, given the error seen. Actually, this draft PR doesn't capture the full changes at the moment. Look at PR #262 to see the full changes.

Details:

1. I put this as draft but in process of trying refactor this draft PR to remove any use of CUDA 11.0.3 in the CI. I will update the title with NOTHING TO SEE HERE. Feel free to comment here, but better would be PR CI test with CUDA 11.0 + Perfetto connector fix (bringing PR 212 to completion) #262.

2. Note that constexpr problem you refer to is not the only fix in CI test with CUDA 11.0 + Perfetto connector fix (bringing PR 212 to completion) #262 that is to come here - the primary problem with perfetto-connector is BUILTIN_CLOCK.

3. The problem remaining for CUDA 11.0.3 is with nvcc_wrapper - see the CI test for CUDA 11.0.3 in CI test with CUDA 11.0 + Perfetto connector fix (bringing PR 212 to completion) #262. I think there is a problem involving the docker container - my goal is to make that a separate GitHub PR so that the other changes made in PR CI test with CUDA 11.0 + Perfetto connector fix (bringing PR 212 to completion) #262 can be merged.