[DNM] [rebranch] Correct nullability of free() redeclaration

[beccadax]

In the new clang, this declaration ends up influencing the redeclaration chain, altering the nullability of the free(_:) function imported into Swift.

Fixes rdar://71269128. Note that tests may fail unless a clean build is performed.

I'm not yet certain whether I will merge this to rebranch or to main, but I know that I want a PR test against rebranch.

[beccadax]

@swift-ci please smoke test

[beccadax]

19:35:26 
/home/buildnode/jenkins/workspace/swift-PR-Linux-smoke-test@2/branch-rebranch/swift/lib/IRGen/GenCall.cpp:496:35: error: no member named 'SwiftAsync' in 'llvm::Attribute'
19:35:26   b.addAttribute(llvm::Attribute::SwiftAsync);
19:35:26                  ~~~~~~~~~~~~~~~~~^
19:35:26 /home/buildnode/jenkins/workspace/swift-PR-Linux-smoke-test@2/branch-rebranch/swift/lib/IRGen/GenCall.cpp:1538:33: error: no member named 'SwiftAsyncContext' in 'clang::ParameterABI'; did you mean 'SwiftContext'?
19:35:26       case clang::ParameterABI::SwiftAsyncContext:
19:35:26            ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~
19:35:26                                 SwiftContext
19:35:26 /home/buildnode/jenkins/workspace/swift-PR-Linux-smoke-test@2/branch-rebranch/llvm-project/clang/include/clang/Basic/Specifiers.h:347:5: note: 'SwiftContext' declared here
19:35:26     SwiftContext
19:35:26     ^
19:35:26 /home/buildnode/jenkins/workspace/swift-PR-Linux-smoke-test@2/branch-rebranch/swift/lib/IRGen/GenCall.cpp:1541:12: error: duplicate case value 'SwiftContext'
19:35:26       case clang::ParameterABI::SwiftContext:
19:35:26            ^
19:35:26 /home/buildnode/jenkins/workspace/swift-PR-Linux-smoke-test@2/branch-rebranch/swift/lib/IRGen/GenCall.cpp:1538:12: note: previous case defined here
19:35:26       case clang::ParameterABI::SwiftAsyncContext:
19:35:26            ^

Failure is unrelated.

[beccadax]

Added a revert commit which should work around that failure.

@swift-ci please smoke test

[beccadax]

@swift-ci please smoke test

[beccadax]

We're hitting unrelated bugs in the LLDB configuration, which prevent us from getting useful results from PR tests.

In local testing, I observed that:

• stdlib/TestData.swift passes
• llbuildSwift still fails to compile, but it fails by crashing in IRGen, not by a use of free(_:) failing to typecheck.

On this basis, I believe this change is good and should be merged to main branch.