Fix thrust namespace for CUDA 12.6

[miwechner]

Affecting compilation issue mentioned here: #417 (comment). The other device-code compilation errors with CUDA 12.5 mentioned in the issue are not adressed with this PR.

Thrust changed the namespace for its implementations from thrust:: to thrust::<<ABI_IDENTIFIER>>:: since thrust 2.3.0.

For the few template specializations in iterator_impl, its necessary to to use the correct namespaces. Using the THRUST_NAMESPACE_BEGIN/END marcos should make sure that we always use the correct namespace, even with older CUDA Toolkit versions, but I did not exhaustively test that.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.34%. Comparing base (3b7d712) to head (4d6bb20).
Report is 5 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #428   +/-   ##
=======================================
  Coverage   97.34%   97.34%           
=======================================
  Files          32       32           
  Lines        2524     2524           
=======================================
  Hits         2457     2457           
  Misses         67       67           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[stotko]

Thanks for working on a fix for CUDA 12.6. For reference, the respective namespace macros were introduced in thrust 1.13.1 and CUDA 11.5. We technically also support older CUDA version down to 11.0, so a fallback would actually be needed for those. However, I do not believe that such old versions are still in use, so just bumping the minimum requirements should be sufficient. I will handle the respective documentation changes. Other than that, LGTM. Thanks again for the contribution!