feat: make inlay hints insertable
Part of #13812
This PR implements text edit for inlay hints. When an inlay hint contain text edit, user can "accept" it (e.g. by double-clicking in VS Code) to make the hint actual code (effectively deprecating the hint itself).
This PR does not implement auto import despite the original request; text edits only insert qualified types along with necessary punctuation. I feel there are some missing pieces to implement efficient auto import (in particular, type traversal function with early exit) so left it for future work. Even without it, user can use `replace_qualified_name_with_use` assist after accepting the edit to achieve the same result.
I implemented for the following inlay hints:
- top-level identifier pattern in let statements
- top-level identifier pattern in closure parameters
- closure return type when its has block body
One somewhat strange interaction can be observed when top-level identifier pattern has subpattern: text edit inserts type annotation in different place than the inlay hint. Do we want to allow it or should we not provide text edits for these cases at all?
```rust
let a /* inlay hint shown here */ @ (b, c) = foo();
let a @ (b, c) /* text edit inserts types here */ = foo();
```
Don't suggest unstable items on stable toolchain
Closes#3020
This PR implements stability check in `ide-completion` so that unstable items are only suggested if you're on nightly toolchain.
It's a bit unfortunate `CompletionContext::check_stability()` is spammed all over the crate, but we should call it before building `CompletionItem` as you cannot get attributes on the item it's completing from that struct. I looked up every callsite of `Builder::add_to()`, `Completions::add[_opt]()`, and`Completions::add_all()` and inserted the check wherever necessary.
The tests are admittedly incomplete in that I didn't add tests for every kind of item as I thought that would be too big and not worthwhile. I copy-pasted some existing basic tests in every test module and adjusted them.
fix: when running the "discoverProjectCommand", use the Rust file's parent directory instead of the workspace folder
This is a quick fix to allow the `discoverProjectCommand` to run successfully when the user has a workspace that does not, e.g., have a `.buckconfig` defined.
(It's also probably _more correct_ to set the `pwd` of the command to the parent of the Rust file _anyways_ rather than relying on the workspace folders, which may be entirely unrelated.)
Compute closure captures
This PR:
* Computes closure captures and the trait it implements (Fn, FnMut or FnOnce)
* Computes data layout of closures
* Adds support for closure MIR lowering
* Changes the closure type display from `|arg1: ty1, arg2: ty| -> ret` to `impl FnX(arg1: ty1, arg2: ty2) -> ret`
fix#12297
fix: unify types in `infer_expr_coerce_never()`
Fixes#14506#14506 turned out to be a regression in type inference. `infer_expr_coerce_never()` added in #14251 allows never-to-any coercion but should also perform ordinary type unification in other cases.
