fix(up): stop pinning a CPU core during foreground wait

[Cyb3rDudu]

container-compose up (foreground) holds 100% of one core after all containers have started. Reproducible with any compose file run without -d:

services:
  idle:
    image: alpine:latest
    command: ["sleep", "60"]

$ container-compose up &
$ ps -o pcpu,comm -p $!
%CPU COMM
100.0 container-compose

The hot loop is ComposeUp.swift::waitForever:

for await _ in AsyncStream<Void>(unfolding: {}) {
    // This will never run
}

AsyncStream<Void>(unfolding: () async -> Void?) ends only when the closure returns nil. The empty closure returns (), which Swift auto-wraps as .some(()) — never nil — so the stream emits an infinite sequence of Void values with no await between them. The for-loop iterates as fast as the runtime can deliver them. The // This will never run comment is inverted: the body runs constantly; only fatalError("unreachable") is unreachable.

Fix is withUnsafeContinuation { _ in }: the task suspends until the process is killed, with zero CPU. After this change the same repro shows 0.0 container-compose.

Used withUnsafeContinuation rather than withCheckedContinuation because the latter logs a "continuation leaked" diagnostic at runtime — leaking is the intent here, since the contract is to wait until the process is killed.

Fixes #27.