test: add failure propagation e2e test for odin_test

[clappingmonkey]

Summary

Adds a negative e2e test that validates odin_test correctly propagates non-zero exit codes from failing @(test) procedures to Bazel's test infrastructure.

Architecture

Follows the rules_rust pattern (one workspace per e2e scenario):

e2e/failure_propagation/
  MODULE.bazel              # separate workspace with local_path_override
  BUILD.bazel               # odin_test(manual) + sh_test
  tests_fail/
    failing_test.odin       # @(test) with intentional failure (1+1==3)
  check_exit_code.sh        # runs binary via runfiles, asserts exit != 0

How it works

1. odin_test(name = "should_fail", tags = ["manual"]) — compiled but never auto-run by bazel test //...
2. sh_test(name = "failure_propagation_test") — executes the binary directly (no Bazel-in-Bazel)
3. Uses rlocationpath + runfiles.bash v3 for cross-platform binary path resolution
4. Asserts the binary exits non-zero → PASS; exits zero → FAIL

Cross-platform

• rules_shell v0.8.0 for sh_test (required since Bazel 9.x removed native sh_test)
• @rules_shell//shell/runfiles for portable path resolution
• Works on Linux, macOS, and Windows (GitHub Actions runners have Git Bash)

CI

Adds a new step to the test matrix: "Failure propagation test (e2e)" — runs bazel test //... in e2e/failure_propagation/ across all 4 OS × 3 Bazel combinations.

Verification

• bazel test //... in e2e/failure_propagation: PASSED locally
• pre-commit run --all-files: all checks pass
• Root ruleset bazel build //...: unaffected (.bazelignore updated)

Checklist

• Conventional commit title
• Tests pass locally
• Pre-commit hooks pass
• Follows rules_rust e2e workspace pattern (separate workspace per scenario)