[C++] Temporal floor/ceil/round kernels could be optimised with templating

[CeilTemporal, FloorTemporal, RoundTemporal](https://github.com/apache/arrow/blob/master/cpp/src/arrow/compute/kernels/scalar_temporal_unary.cc#L728-L980) kernels could probably be templated in a clean way. They also execute a switch statement for every call instead of creating an operator at kernel call time and only running that.

**Reporter**: [Rok Mihevc](https://issues.apache.org/jira/browse/ARROW-15787) / @rok
**Assignee**: [Rok Mihevc](https://issues.apache.org/jira/browse/ARROW-15787) / @rok
**Watchers**: [Rok Mihevc](https://issues.apache.org/jira/browse/ARROW-15787) / @rok
#### PRs and other links:
- [GitHub Pull Request #13043](https://github.com/apache/arrow/pull/13043)

_{**Note**: *This issue was originally created as [ARROW-15787](https://issues.apache.org/jira/browse/ARROW-15787). Please see the [migration documentation](https://github.com/apache/arrow/issues/14542) for further details.*}