🍒 [clang] Do not duplicate "EnableSplitLTOUnit" module flag

[kateinoigakukun]

Cherry-pick 53adfa8 to unlock hermetic-seal-at-link on non-Darwin platforms.

Since .bc can be archived into .a without processing it by clang linker, IRGen is responsible to add EnableSplitLTOUnit. However, main module's .bc is processed by clang linker, then EnableSplitLTOUnit is added twice.

module flag identifiers must be unique (or of 'require' type)
!"EnableSplitLTOUnit"
LLVM ERROR: Broken module found, compilation aborted!

---

Original commit message

If clang's output is set to bitcode and LTO is enabled, clang would
unconditionally add the flag to the module. Unfortunately, if the input were a
bitcode or IR file and had the flag set, this would result in two copies of the
flag, which is illegal IR. Guard the setting of the flag by checking whether it
already exists. This follows existing practice for the related "ThinLTO" module
flag.

Differential Revision: https://reviews.llvm.org/D112177

[kateinoigakukun]

@swift-ci Please test

[MaxDesiatov]

Hi @compnerd could you review this PR? It's a cherry-pick that's going to unblock one more PR in the Swift toolchain. Thanks!

[compnerd]

I'm curious about the ignored verifier. Why does the verifier seem to be unhappy here?

[kateinoigakukun]

@compnerd As far as I confirmed, the test case passed even without the -disable-llvm-verifier option. I'm not sure the reason why the verifier can be unhappy in some situations, but I kept the lines as the original upstream patch does.

[compnerd]

Keeping in sync with upstream sounds good

CC @hyp @bnbarham

[MaxDesiatov]

Hi @compnerd, can you merge this PR? Neither I nor @kateinoigakukun have permissions to merge in this repo. Thank you.