Fix CLAUDECODE flag to properly suppress test output

Fixed two issues with the CLAUDECODE environment variable:

1. In `isAIAgent()`, when AGENT was set to any value other than "1"
   (like "false" in test harness), it would return early without
   checking CLAUDECODE. Changed to only return true if AGENT="1",
   otherwise continue checking other environment variables.

2. In `writeTestStatusLine()`, restored the check to suppress
   pass/skip/todo output when running under AI agents. This check
   was removed during a previous refactoring.

Now when CLAUDECODE=1, test output only shows failures, making it
easier for AI agents to focus on what needs attention.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

src/cli/test_command.zig

@@ -62,6 +62,11 @@ fn fmtStatusTextLine(status: bun_test.Execution.Result, emoji_or_color: bool) []
 }
 
 pub fn writeTestStatusLine(comptime status: bun_test.Execution.Result, writer: anytype) void {
+    // When using AI agents, only print failures
+    if (Output.isAIAgent() and comptime status.basicResult() != .fail) {
+        return;
+    }
+
     switch (Output.enable_ansi_colors_stderr) {
         inline else => |enable_ansi_colors_stderr| writer.print(comptime fmtStatusTextLine(status, enable_ansi_colors_stderr), .{}) catch unreachable,
     }

src/output.zig

@@ -463,7 +463,9 @@ pub fn isAIAgent() bool {
         var value = false;
         fn evaluate() bool {
             if (bun.getenvZ("AGENT")) |env| {
-                return strings.eqlComptime(env, "1");
+                if (strings.eqlComptime(env, "1")) {
+                    return true;
+                }
             }
 
             if (isVerbose()) {