fix(ci): limit merge-queue unit tests to ubuntu runners

[kakkoyun]

Summary

• The merge queue's check response timeout is 60 minutes ("Main branch" ruleset), but the unit-test matrix on a merge group is paced by the Windows legs, which take about 55 minutes each (3301-3310s measured on a completed queue batch). The "Complete" check only reports once the slowest leg concludes, at 58-65 minutes, so entries land on both sides of the deadline.
• Since the evening of Sep 15, several fully green batches (fix(pin): avoid inconsistent vendoring failure in pruneImports #886, fix(pin): sync the enclosing go.work with the module's go directive #902, release: v1.13.1 #903) were evicted by github-merge-queue[bot] at about the +60 minute mark. Today fix(pin): sync the enclosing go.work with the module's go directive #902 merged with its checks concluding at +62 minutes, while release: v1.13.1 #903 was evicted in the same minute with its checks concluding at +67.
• This change runs only the ubuntu legs (about 28 minutes) of the unit-test matrix on merge_group events, bringing the whole batch to roughly half an hour, comfortably inside the timeout. Pull request and push runs keep the full ubuntu + windows matrix, and the E2E, integration test, and lint jobs continue to cover Windows on merge groups.
• Not the long pole, for the record: benchmarks finish in about 11 minutes on merge groups, so they are left unchanged. The ruleset timeout itself is not code and needs a repo admin to change.

Test plan

• On a pull request: all four unit-test legs (ubuntu/windows x stable/oldstable) still run.
• In the merge queue: only the ubuntu legs run, and the batch completes in about half an hour.

[datadog-datadog-us1-prod]

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 75.61%

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 7ad94fe | Docs | View more details | Give us feedback!}