chore(compiler): bump amd-llvm to SMP 23.11 (hotswap sync to amd-staging)

[lamb-j]

Summary

Bumps compiler/amd-llvm from 9f1cce5a to 8bbfd71, bringing the comgr hotswap content up to parity with amd-staging.

The prior pin was behind amd-staging on comgr hotswap. These nine commits are cherry-picked on top of the current pin, in amd-staging landed order:

PR	Title
—	[comgr][lit] Need gfx1250
ROCm/llvm-project#3032	[Comgr] Fix hotswap-rewrite-e2e lit test
ROCm/llvm-project#3015	[Comgr][NFC] Disable CMake EXPORT_COMPILE_COMMANDS within HotSwap
ROCm/llvm-project#3065	comgr: Use triple form of lookupTarget
ROCm/llvm-project#3098	AMDGPU: add hotswap entry trampoline lit coverage
ROCm/llvm-project#3417	[comgr][hotswap] Place entry-stub symbols in the trampoline pool, not .text
ROCm/llvm-project#3454	[Comgr] Fix build due to missed signature update
ROCm/llvm-project#3496	[comgr][hotswap] use global_prefetch_b8 for GFX1250 unclaused-VMEM entry workaround
ROCm/llvm-project#3428	[comgr][hotswap] Preserve VGPR-MSB mode when splitting gfx1250 WMMAs

Each carries a (cherry picked from ...) provenance line. Two needed trivial resolution:

• Compiler Update LLVM Submodules Feb 13 #3417 — its only conflict was a whitespace-only edit inside unpackage(), a function absent from this branch; dropped (kept HEAD).
• [Multi-arch] Add MIOpen database handler for kpack artifact splitting #3496 — conflicted on two lit .s CHECK files; took amd-staging's versions.

After the picks, the comgr hotswap sources, the hotswap/ transpiler dir, and all hotswap lit + unit tests are byte-identical to amd-staging, with one intentional exception (follow-up commit): the gfx12-5-generic RUN lines are dropped from hotswap-kernel-entry-trampoline-gfx125x.s. That variant exercises the gfx12-5-generic entry-trampoline rewrite, which needs gfx12-5-generic assembler support newer than this pin's LLVM base and returns INVALID_ARGUMENT; gfx1250 and gfx1251 both pass. The coverage returns when the LLVM base advances.

Not included (not hotswap)

General comgr divergence from the pin's older base — exportmap.in (unpackage/package-entry ABI) and README.md (AMD_COMGR_LOG_LEVEL, AMD_COMGR_USE_EMBEDDED_LIBCXX docs) — left out of scope.

Numbering

Eleventh compiler-pin change since the ww23 weekly landed (the current pin 9f1cce5a is the tenth), so this is SMP 23.11.

ISSUE ID: #6235

[therock-pr-bot]

✅ All Policy Checks Passed

Check	Status	Details
🌿 Branch Name	✅ Pass	—
📝 PR Title/Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	✅ Pass	PR does not contain code files — Unit Test auto-passed
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—

🎉 All policy checks passed!

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

[therock-pr-bot]

🎉 All checks passed! This PR is ready for review.

[lamb-j]

Now that the compiler-runtime tests are blocking, and we have the Comgr Gtests turned on, we should get a nice clear CI signal for a failure instead of needing to dig