CLI: propagate alias rename errors, doctor config resilience

aliases: rename now propagates every meta.json I/O and parse error
instead of silently ignoring failures. Previously, a corrupt or
unreadable meta.json after directory rename would leave the alias
in an inconsistent state with no user-visible error.

doctor: Config::load failure now falls back to Config::default()
with a warning instead of aborting the entire health check. Doctor
should still run diagnostics even when config is missing or corrupt
— that's exactly when you need it most.

src/cli/cache_cmd.rs

@@ -32,8 +32,8 @@ pub struct Args {
     #[arg(long)]
     pub prune_stale: bool,
 
-    /// Days before an alias is considered stale (default: 90)
-    #[arg(long, default_value_t = 90)]
+    /// Days before an alias is considered stale (default: 30, matching config)
+    #[arg(long, default_value_t = 30)]
     pub prune_threshold: u32,
 
     /// Evict least-recently-used aliases until total size is under this limit (MB)
@@ -290,13 +290,13 @@ fn execute_prune(args: &Args, robot: bool, start: Instant) -> Result<(), Swagger
             "cache",
             start.elapsed(),
         );
-    } else if stale.is_empty() {
+    } else if pruned.is_empty() {
         println!(
             "No stale aliases (threshold: {} days).",
             args.prune_threshold
         );
     } else {
-        println!("Pruned {} stale alias(es).", stale.len());
+        println!("Pruned {} stale alias(es).", pruned.len());
     }
     Ok(())
 }