feat: upgrade llama.cpp to b11018

[bernardladenthin]

Summary

• Upgrades llama.cpp b11012 → b11018 — 6 commits, 28 KiB, 9 files. Under the runbook's 100 KiB threshold, so a single step with no chunking.
• Eight of the nine files are ggml/src backend internals; the ninth is CODEOWNERS. Zero files under common/, include/, tools/server/ or tools/mtmd/, and no patch target touched.
• All nine patches untouched; all six standing drop-checks still report still required.

What changed

Commit	Area
#28953	SYCL — fixes a B70 allocation failure above 19.3 GB
#28929	SYCL — fuses the SiLU epilogue into the ssm_conv kernel (new ssm_conv.{cpp,hpp}, fusion.cpp)
#25483	Vulkan — skips unneeded MoE work in the mul_mm coopmat1 path
#28996	Vulkan — fixes buffer_reference alignment in im2col.comp / im2col_3d.comp
#28984	OpenCL — clears various warnings
#29003	docs — removes a code owner for test-llama-archs

Every change is confined to a backend the classifier jobs build but whose internals this project never calls; the default JAR's CPU path is untouched. Every row of the API-compatibility table is vacuously satisfied and the three mechanical tools/server contract greps have no input.

Test plan

• Affected unit / integration tests pass locally
• CI is green on this branch — see the note below
• Docs / CHANGELOG updated where applicable — two rows appended to docs/history/llama-cpp-breaking-changes.md

Verified at the target tag from a fresh configure (rm -rf build, real FetchContent path):

Check	Result
Configure	clean; stamp head c9a5eeeb3 with nine SHA-256 lines
verify-patches-applied.sh	9 applied, tree dirty, patches/0010 cast present
Wire-name extraction	138 CLI / 57 request / 15 trainer — unchanged
Release build	clean, 0 errors, 0 warnings
ctest	551/551
nm -D	40 Java_* exports, 0 C++-mangled
NativeLibraryLoadSmokeTest	4/4, 0 skipped after a clean — cross-validates LLAMA_CPP_VERSION against the linked build-info
mvn test	1763 tests, 0 failures
SpotBugs / spotless	0 bugs, clean

All six standing drop-checks against the pristine tag — all still required:

0001 (common_params_parse_main 0 occurrences in b11018:common/arg.h, WIN32 override still at common/arg.cpp:1282) · 0002 (load_progress_callback still unguarded at server-context.cpp:1095) · 0003/0006/0008 (all three symbols absent from b11018:tools/server/) · 0010 (vocab_type still uncast at server-context.cpp:4554) · 0012 (bare splits[i] /= split_sum; still at src/llama-model.cpp:1518, unmoved from b11012)

Note

The previous PR (#439) got a real full-matrix run — #958, 2h20m, 66 jobs: 58 success / 2 failure / 6 skipped. The two failures are the Verify GPG signing key pair, which publish.yml documents as the expected red on a pull_request event (the maven-central environment withholds secrets there).

That run was the first full-matrix execution since b10948, and it is what finally exercised the CI wiring added earlier in this series: the trainer-model download, verify-test-counts.sh on all six Java jobs, and both aarch64 fat-jar smoke jobs — all green.

This PR's own run is subject to the usual start-gate abort window, so it may or may not execute; the same two GPG jobs will be red either way.

Related issues / PRs

Follows #439. No upstream issue — nothing in this range required a report.

Checklist

• I have read CONTRIBUTING.md and CODE_OF_CONDUCT.md
• My commits follow Conventional Commits
• No security-sensitive changes

🤖 Generated with Claude Code

https://claude.ai/code/session_01Cft7guQngfyKycdJfBEfJP

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud