Add column_accumulate and generalize column_reduce

[dennisYatunin]

This PR adds a column_accumulate! function that can accumulate values along each column of a Field, a pointwise Broadcasted object, or a columnwise StencilBroadcasted object. It also generalizes column_mapreduce!, which could only accept Fields, to column_reduce!, which can accept Broadcasted and StencilBroadcasted objects as well.

In addition, the column_integral_definite! and column_integral_indefinite! functions have been redefined in terms of column_reduce! and column_accumulate!, respectively. This API will generally allow us to define arbitrary "vertical sweep" operations without the need for custom kernel functions or external caches, which should allow us to simplify the non-orographic gravity wave parametrization in ClimaAtmos.

This PR also updates ClimaComms to 0.6.4 in all Manifest files to make use of the new device-flexible assert macro.

• Code follows the style guidelines OR N/A.
• Unit tests are included OR N/A.
• Code is exercised in an integration test OR N/A.
• Documentation has been added/updated OR N/A.

[charleskawczynski]

Looks good to me, I left some comments, my main comment is: can we add some more granular unit tests?

[dennisYatunin]

I've enabled GPU support and added more unit tests. Adding the error handling for find_zero actually revealed that one of the profiles in ClimaAtmos was not converging for the default absolute tolerance of 0.001, so I changed it to a relative tolerance of 0.001. I'm going to move the device-specific @assert to ClimaComms, and then this will be ready to merge in. @charleskawczynski, could you please take another look?