[5.9] Fix #64757: Unexpected "Implicit use of 'self' in closure" error in closure nested in result builder

[calda]

This PR cherry picks #64786 into the Swift 5.9 release branch.

• Explanation: This fixes closure capture rule applied at Swift 5.8 doesn't work correctly in resultBuilder #64757, where shorthand guard let self and if let self conditions would unexpectedly be rejected inside result builders.
• Scope: This fixes a source-break accidentally introduced by Swift 5.8
• Issue: fixes closure capture rule applied at Swift 5.8 doesn't work correctly in resultBuilder #64757
• Risk: There is somewhat limited risk here since this change was merged to master just three weeks after 5.9 branch cut, and because the change itself is small. This change updates a flag related to handling code completion (this flag was first added in 2c5b193). The worst case risk of this CP would be returning the code-completion experience back to what it was before 2c5b193.
• Testing: CI test suite and SourceKit stress tests
• Reviewers: @DougGregor (5.9 release manager), @xedin (original reviewer of [SE-0365] Fix #64757: Unexpected "Implicit use of 'self' in closure" error in closure nested in result builder #64786)

---

Example of #64757:

@resultBuilder
struct VoidBuilder {
    static func buildBlock() -> Void { }
    static func buildPartialBlock<T>(first: T) -> Void { }
    static func buildPartialBlock<T>(accumulated: Void, next: T) -> Void { }
}

final class EscapingWrapper {
    static func wrapper(_ closure: @escaping () -> Void) {
        closure()
    }
}

final class Sample {
    @VoidBuilder
    var void: Void {
        EscapingWrapper.wrapper { [weak self] in
            guard let self else { return } // Unexpected error: Implicit use of 'self' in closure; use 'self.' to make capture semantics explicit
        }
    }
}

[xedin]

@calda Please use template for description outlined in https://forums.swift.org/t/swift-5-9-release-process/63557

[calda]

Updated @xedin

[xedin]

@calda I don’t think this regression is in 5.8, is it? Alex’s changes are only merged to main.

[calda]

@xedin This is a problem in 5.8, as reported by #64757. The sample code above compiles successfully in 5.7.3 but not 5.8 or 5.9.

You're right that the issue in 5.8 wasn't caused by Alex's change in 2c5b193. My mistake. I'm actually not sure what caused this regression in Swift 5.8, but it must have been something else.

[xedin]

I see, I think it might be caused by the same conceptual problem because closures are still “type-checked separately” (typeCheckClosureBody is called although solver did include the body in inference) during solution application process, we noticed this too late to fix for 5.8.

[xedin]

@swift-ci please test

[calda]

Closing since Swift 5.9 is pretty close to being released and this CP doesn't seem like it would qualify anymore. Fix will be included in 5.10.