fix: Closure capturing for let exprs

[ShoyuVanilla]

Fixes #18201

We have been handling closure captures for let expressions in a wrong way:

Here,

rust-analyzer/crates/hir-ty/src/infer/closure.rs

Lines 1237 to 1241 in 6807f60

	| Expr::Let { pat: _, expr }
	| Expr::Box { expr }
	| Expr::Cast { expr, type_ref: _ } => {
	self.consume_expr(*expr);
	}

which ends up as a capture by value if the captured target is not Copy

rust-analyzer/crates/hir-ty/src/infer/closure.rs

Lines 1034 to 1051 in 6807f60

	fn consume_expr(&mut self, expr: ExprId) {
	if let Some(place) = self.place_of_expr(expr) {
	self.consume_place(place);
	}
	self.walk_expr(expr);
	}
	fn consume_place(&mut self, place: HirPlace) {
	if self.is_upvar(&place) {
	let ty = place.ty(self);
	let kind = if self.is_ty_copy(ty) {
	CaptureKind::ByRef(BorrowKind::Shared)
	} else {
	CaptureKind::ByValue
	};
	self.push_capture(place, kind);
	}
	}

But rustc handles this as a capture by ref(immutable, in most cases) in the following lines:

https://github.com/rust-lang/rust/blob/6f935a044d1ddeb6160494a6320d008d7c311aef/compiler/rustc_hir_typeck/src/expr_use_visitor.rs#L452-L454
https://github.com/rust-lang/rust/blob/6f935a044d1ddeb6160494a6320d008d7c311aef/compiler/rustc_hir_typeck/src/upvar.rs#L2093-L2114

Of course, rustc's capturing logic is more sophiscated than my modification here and we lack many details of them but at least, this one is at least more precise than the present, I think 😅

I think that we should move our capturing logic closer to that of rustc, but after next-gen solver migration