Fix #7193: persistently remember what is a projection

[andreasabel]

Introduces a new FunctionFlag FunProj that remembers whether this function is or came from a projection.
This flag is more stable than the funProjection data which relies on arity.
This helps ruling out accessing irrelevant fields in irrelevant definitions without activating --irrelevant-projections.

Fixes #7193. (Hopefully, feels almost too good to be true.)

[andreasabel]

@dolio: Please review, if you are available.

[dolio]

It seems good to me. I think this is addressing the root cause, which was that Agda was losing track of which things are projections. I wasn't the one who originally came up with either trick, so it's possible that there are others I don't know about. But it fixes the ones I do.

I noticed that when you do the let open TSquash (emb x) in extract example, extract is still colored like a normal function (when irrelevant projections are on, or the projections aren't irrelevant). Is it possible to use this new information to color it pink like they are within the record, or when you open an unapplied record module? Seems like that'd be nice.

[andreasabel]

Thanks @dolio!

I noticed that when you do the let open TSquash (emb x) in extract example, extract is still colored like a normal function (when irrelevant projections are on, or the projections aren't irrelevant). Is it possible to use this new information to color it pink like they are within the record, or when you open an unapplied record module?

@nad requested the opposite, if I am not mistaken:

• Do not highlight projection functions in expressions as fields #4604

But maybe we can fix the following problem now: #3156.
(No, we can only make things more pink, not less.)