ci: isolate native ABI smoke testing

[bernalde]

Summary

• move the native C ABI build and two C-client executions into one dedicated Ubuntu job
• keep the pinned upstream MQLib revision unchanged
• disable unused coverage instrumentation in all three Julia compatibility jobs

Validation

• actionlint
• bash test/native_c_api_smoke.sh
  • built upstream commit 5854962
  • executed solve_qubo and c_api_smoke successfully
• git -c core.whitespace=blank-at-eol,blank-at-eof,space-before-tab,cr-at-eol diff --check
• full CI matrix (post-push; timings to be recorded from Actions)

Measurement

The #34 baseline ran the identical 25-second native smoke on both Ubuntu matrix jobs. This PR runs it once in parallel without Julia setup.

• attempt 1: 10.4 aggregate runner-minutes; dedicated native smoke 30s
• warm rerun: 8.3 aggregate runner-minutes; dedicated native smoke 32s
• all Julia jobs logged coverage=false; the native ABI test executed exactly once per workflow run

The warm Windows job varied upward (3m45s to 4m45s), so the aggregate figures should not be treated as a deterministic benchmark. The structural saving is one fewer upstream clone/build and two fewer native client executions per workflow run.

Branch Hygiene

• Base: main at 5664fcd2b943343ee9fad2aa7fd93d0b1743b468
• Not stacked; open PR Make compatibility checks Dependabot-safe #33 changes test/runtests.jl, while this PR changes only the CI workflow.
• Existing mixed line endings in ci.yml were preserved; whitespace validation treats CR at EOL correctly.

Closes #34