ci: require the live suites in each platform check

[cv]

Failure

Since #12962, Live / Docker and Live / Kind run inside each Linux platform's reusable workflow with continue-on-error. A live failure never fails Test / PLATFORM, but each gate still waits for the platform's whole workflow, live jobs included. Required checks therefore wait for the live suites without depending on their result.

In run 38070042632, measured from when its jobs started, every check that decides the gates finished by 674 s. Linux AMD64 Live / Docker finished at 1,077 s, and the gates at 1,080 s.

Decision

Remove continue-on-error from both live jobs, so a live failure fails its platform's Test / PLATFORM check, as a lifecycle failure does. When a change cannot affect the live suites, each live job skips its remaining steps and passes, as before. Check names are unchanged.

The suites have been reliable enough to require. Since #12962, Linux Docker and Kind ran 19 times each and passed every run that wasn't cancelled: 34 of 34. The six other jobs were all cancelled.

When the live suites run, they still set how long required checks take: Live / Docker on Linux AMD64 has a median of 810 s and a maximum of 1,524 s. Follow-up changes will cache Docker layers between runs and reuse the test executables Build and test already compiles.

Validation

• actionlint passes on native-platform.yml and rust.yml.
• cargo ci passed: 1,326 workspace tests and 128 lifecycle tests.
• This PR's own run exercises the live jobs, because it changes native-platform.yml.

[github-actions]

v1 documentation preview: https://nvidia-preview-nemoclaw-v1-pr-12970.docs.buildwithfern.com/nemoclaw