add compute kernels that operate on `ArrayRef` -- e.g `eq_dyn`

[alamb]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.
I want to compare two ArrayRefs (Arc<Array>) with each other

Arrow offers compute kernels for primitive (numeric) types(e.g. eq), and separate ones for string types (e.g. eq_utf8).
https://docs.rs/arrow/6.0.0/arrow/compute/kernels/comparison/index.html

This means I have to match on the datatype and then dispatch to the appropriate implementation

You can see a bunch of this kind of dispatch (encoded in macros) in Datafusion, for example https://github.com/apache/arrow-datafusion/blob/81fae230b81b97e93bbb95b284cfda6f4d59552e/datafusion/src/physical_plan/expressions/binary.rs#L289-L328

Describe the solution you'd like
Add functions like OP_dyn that did the type dispatch within arrow

So for example, a function like this:

/// Compare the left and right arrays for equality. Returns an error if they are not the exact same type
fn eq_dyn(left: &dyn Array, right: &dyn Array) -> result<BooleanArray> {
  // switch on types and call appropriate eq kernel
...
}

and neq_dyn, lt_dyn, etc.

Describe alternatives you've considered
Leave it as is

Additional context
See also apache/datafusion#1163 and #842

[sum12]

I guess the more the better, but this is the complete list ?

[alamb]

I guess the more the better, but this is the complete list ?

I was thinking more like the kernels in arrow itself:
https://docs.rs/arrow/6.0.0/arrow/compute/kernels/index.html

So definitely for the comparison kernels listed
https://docs.rs/arrow/6.0.0/arrow/compute/kernels/comparison/index.html

But also it may make sense for the aggregate kernels https://docs.rs/arrow/6.0.0/arrow/compute/kernels/comparison/index.html

I took a quick look through the rest of the kernels and most of them either already operate on dyn Array (e.g. cast) or they only logically make sense on certain types of arrays (e.g. the Boolean kernels). Though even for boolean kernels it might be convenient to offer a function that does the downcasting appropriately