[ENHANCEMENT]: Improve `make_window_extent` in the case of linear probing

[PointKernel]

Is your feature request related to a problem? Please describe.

The current make_window_extent function always calculates prime capacity, regardless of the probing scheme used. This approach is suboptimal for linear probing, where the capacity does not need to be a prime number.

Describe the solution you'd like

If it's linear probing, return the window extent based on CG size and window size without the prime table lookup.

template <typename T>
struct is_double_hashing : cuda::std::false_type {};

template <int32_t CGSize, typename Hash1, typename Hash2>
struct is_double_hashing<cuco::double_hashing<CGSize, Hash1, Hash2>> : cuda::std::true_type {};

A trait similar to the above idea is needed to dispatch between linear probing and double hashing

Describe alternatives you've considered

No response

Additional context

No response