fix: skip post-job hook when executor fails to boot

[cchristous]

Problem

When a job's executor fails to boot, the agent process crashes with a nil-pointer SIGSEGV instead of failing the job cleanly.

Root cause

In Job.RunWithOptions, the executor is started via PrepareEnvironment(). When that fails, executorRunning is false, so the regular commands and epilogues are correctly skipped — but the post-job hook was called unconditionally afterward:

if executorRunning {
    result = job.RunRegularCommands(options)
    ...
    if !job.Stopped {
        job.handleEpilogues(result)
    }
}

// reached even when executorRunning == false
job.runPostJobHook(options)

runPostJobHook runs the hook inside the executor (Executor.RunCommandWithOptions), which for DockerComposeExecutor/ShellExecutor calls e.Shell.NewProcessWithOutput(...). e.Shell is only assigned once the executor boots successfully, so on a failed boot it is nil and the call dereferences a nil pointer — crashing the whole agent process.

Fix

Gate the post-job hook on executorRunning, matching how the epilogues directly above it are already handled. When the executor never booted, the hook is skipped and the job finishes cleanly as failed (with the existing Failed to prepare executor / Executor failed to boot up log lines).

Test

Adds Test__PostJobHookSkippedWhenExecutorFailsToPrepare: a stub executor whose Prepare() fails drives RunWithOptions with a post-job hook configured, and the test asserts the hook is never attempted (and the job still finishes). The test fails without the guard and passes with it. The existing Test__UsePostJobHook continues to cover the success path.