Possible miscounting of `inv_owned`

[timholy]

The script reports the total number of invalidations as well as all of those "owned" by the package's dependencies:

julia-invalidations/action.yml

Lines 45 to 47 in 610afc0

	inv_owned = length(filtermod(${{ steps.info.outputs.packagename }}, invalidation_trees(invalidations)))
	inv_total = length(uinvalidated(invalidations))
	inv_deps = inv_total - inv_owned

However, I'm unsure what the counting of "owned" items is intended to accomplish. Currently this counts the number of independent trees that contain one or more top-level children owned by the package (i.e., direct invalidations of package code). Unless some additional packages or additional methods were defined after loading the package, I think that number will always be zero, because we don't cache invalidated CodeInstances during precompilation.

My best guess is that this is instead intended to count the number of invalidations caused by methods in the package? In which case something like

sum(SnoopCompile.countchildren, filter(tree -> tree.method.module == ${{ steps.info.outputs.packagename }}, invalidation_trees(invalidations)))

might be more appropriate?