Distribution timing timeout can bypass cleanup with inherited errexit

[n0tl3ss]

PR #1915 review4200719999 identifies missing failure-path cleanup in the distribution timing script. A new six-test stdlib reproducer executes the actual Python ttfr.sh with fake curl/docker/native/kill commands; curl advances Bash SECONDS to exercise timeout immediately, without creating containers or building native images.

Before production changes, four tests pass and two fail. Docker and native success clean up once; default Bash timeouts exit nonzero and clean up. However, with POSIX/inherited errexit enabled, both timeout paths exit before normal cleanup and fail the cleanup assertion. This is a reproduced stricter-shell failure, not a claim that default Bash swallows the existing timeout failure. The Java/Groovy/Kotlin sibling scripts share the same cleanup pattern.

Install EXIT cleanup after a successful launch, preserve failed timing status, and clear the trap after normal cleanup. Retain success and failure controls. Keep the fix scoped to the distribution base; no process/image artifacts are needed for reproduction.

[n0tl3ss]

The reproducer was expanded before applying the mirrored fix: nine unittest methods cover 36 fake-command subcases across Python/Java/Groovy/Kotlin, with 12 inherited-errexit cleanup failures on the original scripts. The remaining 24 success/default-Bash timeout controls pass. After adding EXIT cleanup after each successful launch, clearing it after normal cleanup, and returning failed timing status, all 36 subcases pass. Shell syntax checks pass for all four scripts. No actual containers, native images or Java applications were started by the regression harness.