feat(sessions): Add session command feature

[pthurlow]

No description provided.

[claude]

P2 — Signal exit code lost.

s.code() returns None when the child is killed by a signal (e.g. SIGINT from Ctrl-C), so the wrapper exits 1 instead of the conventional 128+signal. Scripts that inspect the exit code (e.g. if [ $? -eq 130 ]) will misbehave.

On Unix, the idiomatic fix is to re-raise the signal so the parent shell sees the process as signal-killed:

Suggested change

	match status {
	Ok(s) => {
	#[cfg(unix)]
	if let Some(sig) = std::os::unix::process::ExitStatusExt::signal(&s) {
	// Re-raise so the parent shell sees a signal-killed exit
	unsafe { libc::raise(sig) };
	}
	std::process::exit(s.code().unwrap_or(1))
	}

If pulling in libc is undesirable, exiting with 128 + signal is the next-best option.

[claude]

P2 (theoretical) — Arg matching is position-independent.

any(|a| a == "sessions") && any(|a| a == "run") matches any hotdata process that has both tokens anywhere in its argv — e.g. hotdata sessions new --name run. In practice none of the other sessions subcommands spawn child hotdata processes, so this can't actually trigger a false positive today. But it's fragile: a future subcommand that accepts free-form string args and spawns children would silently break.

Prefer a positional check on the subcommand slots:

Suggested change

	let name = proc.name().to_string_lossy();
	if name == "hotdata" {
	if proc.cmd().iter().any(|a| a == "sessions")
	&& proc.cmd().iter().any(|a| a == "run")
	if name == "hotdata" {
	let args: Vec<_> = proc.cmd().iter().skip(1).collect(); // skip argv[0]
	if args.first().map(|a| a.as_encoded_bytes()) == Some(b"sessions")
	&& args.get(1).map(|a| a.as_encoded_bytes()) == Some(b"run")
	{
	return Some(pid);
	}
	}