A few starter generic compute kernels

### Describe the enhancement requested

I have been hand-writing a pile of generic kernels for numerical arrays and wanted to know what would be required to contribute them upstream.

So far I have implementations for:
- `public static Array Filter(Array array, BooleanArray mask)`
- `public static BooleanArray Equal(PrimitiveArray lhs, T rhs) where T: struct, INumber`
- `public static BooleanArray Equal(PrimitiveArray lhs, PrimitiveArray rhs) where T: struct, INumber`
- `public static Array Take(Array array, IList indices)`
- `public static Int64Array CastInt64(PrimitiveArray array) where T: struct, INumber`
- `public static Int32Array CastInt32(PrimitiveArray array) where T : struct, INumber`
- `public static FloatArray CastFloat(PrimitiveArray array) where T : struct, INumber`
- `public static DoubleArray CastDouble(PrimitiveArray array) where T : struct, INumber`

I also have a generic visitor for converting `Large*` arrays to their regular counterparts so that they may work with the other machinery in the library. I am less confident that is useful to others, or that it is truly correct.

Is there a particular pattern you would want these to follow or is the notion not suitable for the .NET ecosystem? I am less familiar with .NET as an ecosystem.

Thank you.

The issue provides no files, tests, or entry points to start from; begin by reviewing the proposed kernels and the generic Large-array visitor in the surrounding Arrow .NET APIs. The first milestone is agreement on the contribution pattern and scope, including whether the visitor is correct and which operations belong upstream.

Written by the indexing model from the issue text.