Cancelling NativeCommandExecutor.Execute kills only the shell: work in a subshell, pipeline or child process keeps running and completes its side effects #97

What's wrong

The synchronous NativeCommandExecutor.Execute handles cancellation by calling TryKill(process), which calls process.Kill() — killing only the direct child, /bin/sh or cmd.exe (NativeCommandExecutor.cs L141-L147, [L210-L221](https://github.com/ktsu-dev/Essentials/blob/415fb1b3f8ef3315b19e6485ce33e2a8441ab853/Essentials.CommandExecutors.Native/NativeCommandExecutor.cs#L210-L221)).``

Every command runs inside a shell (L185-L186),`` so whatever the shell started outlives the kill:

• Unix: subshells, pipelines, backgrounded jobs, any multi-command line where sh forks.
• Windows: cmd.exe /c runs every external program as a child, so cancellation almost never stops the actual command.

The orphan also keeps the redirected stdout/stderr pipes open, so each StreamDrain.Abandon() waits out its full DrainAbandonTimeoutMilliseconds (1 s) (L157-L158, [L303](https://github.com/ktsu-dev/Essentials/blob/415fb1b3f8ef3315b19e6485ce33e2a8441ab853/Essentials.CommandExecutors.Native/NativeCommandExecutor.cs#L303));`` the call returns ~2 s late and leaves two background threads blocked until the orphan exits.

#53 says "the sync path already kills the process, so only the async path is affected" — true only when the shell execs the command directly. Its suggested TryKill fallback would carry this gap into the async fix.

Reproduction (verified on Linux, net10.0, built from 415fb1b)

ICommandExecutor ex = new NativeCommandExecutor();
using var cts = new CancellationTokenSource(500);
var sw = Stopwatch.StartNew();
var r = ex.Execute($"(sleep 3; touch {marker}); echo done", null, null, cts.Token);
Console.WriteLine($"sync cancel returned after {sw.ElapsedMilliseconds}ms: exit={r.ExitCode} err=[{r.StandardError}]");
Thread.Sleep(4000);
Console.WriteLine($"marker exists after cancel? {File.Exists(marker)}");

Observed:

sync cancel returned after 2513ms: exit=-1 err=[Operation was cancelled.]
marker exists after cancel? True

Expected: return within about one poll interval of the cancel, and the marker is never created.

CommandExecutor_Sync_Execute_Cancels_A_Running_Process (CommandExecutorTests.cs L224) misses this: it only asserts elapsed time, and on Unix a bare sleep N is exec'd by the shell, so the one process killed is the right one.

Why it matters

The caller is told the command was cancelled while the work continues and performs its side effects (writes, builds, deploys). On Windows this is the normal case.

Suggested fix

• Kill the whole tree: process.Kill(entireProcessTree: true) on every target except netstandard2.1 (available from netcoreapp3.0), with Kill() as the netstandard2.1 fallback behind the narrowest #if.
• Use the same helper in the async cancellation fix for Cancelling NativeCommandExecutor.ExecuteAsync reports "cancelled" but leaves the child process running (the sync Execute kills it) #53 so both paths agree.

Acceptance criteria

• A test cancels Execute on a command whose work runs in a grandchild of the shell (Unix: (sleep N; touch marker); true; Windows: ping/timeout then a marker write), asserting the marker is never created and the call returns well under the drain-abandon timeout.
• The same test passes for ExecuteAsync once Cancelling NativeCommandExecutor.ExecuteAsync reports "cancelled" but leaves the child process running (the sync Execute kills it) #53 is fixed.

Related: #53.