fix(startup): recover from panics in deferred background init

The deferred background-init steps run in a detached goroutine after the
HTTP listener is already accepting connections. An unrecovered panic in any
step (queue seeding, legacy cleanup, scrobble sweep) would crash the entire
live server. Wrap each step in a recover that logs the panic with a stack
and continues to the next step.

Found during PR #21 review.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

cmd/silo/main.go

@@ -15,6 +15,7 @@ import (
 	"os"
 	"os/signal"
 	"path/filepath"
+	"runtime/debug"
 	"strconv"
 	"strings"
 	"syscall"
@@ -1677,7 +1678,15 @@ func main() {
 				if appCtx.Err() != nil {
 					return
 				}
-				step(appCtx)
+				func() {
+					defer func() {
+						if p := recover(); p != nil {
+							slog.Error("deferred startup init step panicked; continuing",
+								"panic", p, "stack", string(debug.Stack()))
+						}
+					}()
+					step(appCtx)
+				}()
 			}
 			slog.Info("deferred startup init completed", "steps", len(backgroundInit), "duration", time.Since(start))
 		}()