Fix test flakiness

[radubahmata]

Summary

• Isolate tests from host Git configuration.
• Replace fixed delays and wall-clock assertions with observable state and behavior.
• Extend ShellCheck coverage to every test script.

Changes

• .github/workflows/ci.yml: lint all scripts under tests/.
• tests/_test_env.sh: provide shared Git configuration isolation.
• tests/test_ci_workflows.sh: verify all test scripts load the shared environment and receive CI lint coverage.
• tests/test_harness.sh: use a fixed hlog timestamp and explicit writable Git configuration for signing tests.
• tests/runtime_signal_trap.sh: wait for the expected commit before sending signals.
• tests/test_drivers.sh: bound reaper and watchdog tests without asserting wall-clock duration.
• tests/test_launch.sh and tests/test_session_end_push.sh: replace local Git isolation with the shared environment.
• Remaining test scripts: load the shared test environment to prevent host Git settings from affecting results.

Self-review checklist

• This PR addresses a single concern (why?). If it covers multiple independent changes, I've split them into separate PRs.

Test plan

• Run ./tests/test_drivers.sh.
• Run ./tests/test.sh --all.
• Run ShellCheck at warning severity on all shell scripts.

[radubahmata]

Note: usually tests are append-only, but the decision to change the ones covered by this PR is backed by failures encountered in other CI/local test runs. The most recent example: https://github.com/protocol-security/claude-swarm/actions/runs/33561114313/job/100033552849 encountered in #113's CI routine.