Improve simple kernel timer tool

[dalg24]

This is good enough for now.

As a follow up I want us to consider using an environment variable (and command-line argument) to select the output format instead of the KOKKOS_TOOLS_TIMER_JSON boolean option. I also want us to reuse the JSON writing code from kp_json_writer.cpp instead of duplicating. And finally we need to add tests for that tool.

Originally posted by @dalg24 in https://github.com/kokkos/kokkos-tools/pull/235#pullrequestreview-1883183008