Fix insert_or_apply

[srinivasyadav18]

This PR cleans up some of the issues occured during merge of #551.

1. propagate the key_eq and probing_scheme from global ref to constructor of shared_memory_ref in insert_or_apply_shmem kernel.
2. Disable init overload of insert_or_apply using sfinae, because cuda::stream_ref is default constructed, this can invoke the init overload even though the user calls no-init overload.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[PointKernel]

/ok to test