fix: reap zombie child processes in container environments

[toller892]

Problem

When soft-serve runs as PID 1 in a container (e.g. Kubernetes) without an init supervisor like tini, orphaned descendant processes are reparented to PID 1 and become zombies. This happens because:

1. Git operations spawn child processes (e.g. git pack-objects, git index-pack)
2. When the parent git process exits, its children are reparented to PID 1 (soft-serve)
3. The Go runtime only tracks children spawned via os/exec, not reparented orphans
4. Without waitpid() calls, these processes accumulate as zombies

In the reporter's Kubernetes environment, this caused 30k+ zombies in under 24 hours, leading to PID exhaustion and node failure.

Fix

Add a periodic zombie reaper goroutine that calls waitpid(-1, WNOHANG) every 10 seconds to clean up any zombie children. The reaper:

• Runs only on Linux (where the PID 1 container issue manifests)
• Is a no-op on other platforms (macOS, Windows)
• Uses golang.org/x/sys/unix.Wait4 (already a dependency)
• Stops cleanly when the server context is canceled
• Logs reaped PIDs at debug level

Changes

• cmd/soft/serve/reap_linux.go — Linux implementation using unix.Wait4
• cmd/soft/serve/reap_other.go — no-op stub for non-Linux platforms
• cmd/soft/serve/serve.go — call reapZombies() during server startup

Testing

go build ./...    # passes
go test ./cmd/soft/serve/  # passes

Fixes #891

[linsein]

I would prefer to modify the Dockerfile to add tini as the PID 1 process.
If unix.Wait4 is called globally, it might intercept signals intended for cmd.Wait, causing cmd.Wait to throw an error.