Limit completions by keeping only the most relevant ones while generating them

[Veykril]

We currently generate all applicable completions which can be overwhelming in certain scenarios #13124. We should eagerly generate the completion scores while building the completions and instead of keeping an unbounded buffer where we push them into, we should only keep the X highest scored completions in the accumulation.

[Lucretiel]

My concern about this is that it means that certain completions that the user did want might be unreachable, if they happen to be an item with a small name that scores low (maybe a global variable from a different crate; PI comes to mind) and can't compete with higher-scoring stuff (presumably variables more locally defined).

[Veykril]

Well, for one we would have this limit be configurable, so people can set it reasonably high that it wont ever hit them (or turn it off altogether). The limit is mainly useful for projects where you have a large amount of local things (think glob importing bindgen stuff for example). But that aside, assuming the limit is something like 100 items and PI ranks the lowest, you'd have to scroll down all those 100 items anyways.

Note that if your issue here is more about auto-importing completions, those are already limited today for perf reasons (and that limit already impacts people, wonder if we have a concrete issue about that?).