JS: expand localFieldStep to use access-paths, and build access-paths in more cases

[erik-krogh]

trafficstars

Gets a TP/FP for CVE-2022-29078.

Latest performance evaluation shows a ~1.5% performance regression.
And an evaluation on default.yml shows about the same performance regression.

I've battled the performance evaluations on this for a while, and now I'm opening this PR even though I'm not that satisfied with the performance.
But I can't find any reason for the performance regression, and the performance regression disappears when I try to pinpoint it.
(E.g. this evaluation showed a performance regression, but that disappeared when I did a single-threaded experiment on the worst offenders).

I think the performance regression is from us finding more flow, and thus lots of predicates get ever so slightly slower.

An evaluation on the meta queries show some nice new call-edges.