Unexpected behaviour with inverse fieldmask

[will-isovalent]

I noticed this behaviour when constructing an inverse fieldmask that I would consider to be unexpected. Suppose for example we have the following:

fieldmask_utils.MaskInverseFromPaths([]string{"a.b.c","a"}, func(s string) string { return s })

I would expect the resulting fieldmask to be a because the instruction to ignore the entirety of a should take precedence over the instruction to ignore a.b.c. Instead, we get a{b{c}}.