Productize readable kernel names for CI benchmark regression testing

[petebachant]

We want to print out some kernel times and do some comparision with reference values to detect if performance has regressed so deeper analysis can be done.

We want to have a Dict to map object ID to kernel name (computed from stack trace) in auto_launch so we don't need to compute it every time a kernel is launched.

[petebachant]

@imreddyTeja's idea to not need to change GPUCompiler.jl using the cached compiled function

@dennisYatunin's idea to put this in ClimaComms as a Dict to map object IDs to names, start in ClimaCore

IDDict is the correct data type for storing this?