Fix ICE in missing_const_for_fn

[samueltardieu]

The mir_drops_elaborated_and_const_checked query result has been stolen already and cannot be borrowed again. Use the optimized_mir query result instead.

changelog: [missing_const_for_fn]: fix ICE with some compilation options

Fixes #14774

r? @Jarcho

[samueltardieu]

@blyxyas I'm not setup to do perf checks right now, don't hesitate if you want to comment on this change.

[y21]

For what it's worth, this actually used to use optimized_mir and was changed in #14003, so this is effectively a revert of that PR. I do agree though (as I've commented on that PR) that using mir_drops_elaborated_and_const_checked is too fragile and would sooner or later run into random ICEs, because it relies on the assumption that nothing else in the compiler/clippy is using MIR queries before that lint runs (presumably -Zvalidate-mir breaks that assumption), and I agree we should probably should keep using optimized_mir until we have a better solution for #6080.

[blyxyas]

It happened. It really happened that #14003 caused an ICE. I did black magic to try and perform a top 800 lintcheck run on that PR and it was all for nothing. That PR got delayed like 2 months for nothing.

At least it was Matthias who found it and not a random user.

As soon as this is merged, it should probably be backported.

[samueltardieu]

@y21 @blyxyas Feel free to merge it and beta-nominate it, I assigned Jarcho just because when we talked about MIR during the latest meeting he joked that he ended up with all the MIR related PRs.

I'm not sure it deserves to be backported to stable even if there is a 1.87.1, as this is a nursery lint, and seems to only trigger with -Zvalidate-mir so far.

[samueltardieu]

In fact the ICE hadn't entered 1.87 since #14003 was merged some days after the last Clippy sync preceding the 1.87 beta branch creation. Backporting to beta will be enough indeed.