feat: show first-token latency, decode tok/s, per-key colors

- logs page adds First (first_token_ms) and Tok/s columns; Tok/s is
  output_tokens / (latency - first_token) over the decode window
- detail line shows the same rate after first-token latency
- API key names render in a stable per-key color (crc32 over a 10-color
  palette) so rows from the same key group visually; --once stays plain
- --once --logs prints the new first/tok-s columns

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

tests/test_payload.py

@@ -432,6 +432,42 @@ class Sub2APILogsTests(unittest.TestCase):
         capped = mod.logs_request_url("https://sub2apicn.shujk.top/api/v1/admin/usage", 9000)
         self.assertIn("page_size=1000", capped)
 
+    def test_log_tokens_per_second_uses_decode_window(self) -> None:
+        mod = load_module()
+
+        item = {"output_tokens": 340, "duration_ms": 5321, "first_token_ms": 800}
+        self.assertAlmostEqual(mod.log_tokens_per_second(item), 340 * 1000.0 / (5321 - 800))
+        self.assertEqual(mod.format_rate(mod.log_tokens_per_second(item)), "75.2/s")
+        # missing first token: falls back to the full duration
+        self.assertAlmostEqual(mod.log_tokens_per_second({"output_tokens": 100, "duration_ms": 2000}), 50.0)
+        # no output, zero duration, or first token >= duration -> no rate
+        self.assertEqual(mod.log_tokens_per_second({"output_tokens": 0, "duration_ms": 2000}), 0.0)
+        self.assertEqual(mod.log_tokens_per_second({"output_tokens": 10, "duration_ms": 0}), 0.0)
+        self.assertEqual(mod.log_tokens_per_second({"output_tokens": 10, "duration_ms": 500, "first_token_ms": 500}), 0.0)
+        self.assertEqual(mod.format_rate(0), "-")
+        self.assertEqual(mod.format_rate(123.4), "123/s")
+
+    def test_normalize_log_rows_carries_first_token_and_rate(self) -> None:
+        mod = load_module()
+
+        rows = mod.normalize_log_rows(sample_logs_payload())
+        oldest = rows[-1]
+        self.assertEqual(oldest["first_token_ms"], 800)
+        self.assertAlmostEqual(oldest["tokens_per_second"], 340 * 1000.0 / 4521)
+        self.assertIn("first 800ms", mod.log_detail_line(oldest))
+        self.assertIn("75.2/s", mod.log_detail_line(oldest))
+
+    def test_key_color_is_stable_and_from_palette(self) -> None:
+        mod = load_module()
+
+        first = mod.key_color("codex-main")
+        self.assertIn(first, mod.KEY_COLOR_PALETTE)
+        self.assertEqual(first, mod.key_color("codex-main"))
+        self.assertEqual(mod.key_color(""), "")
+        self.assertEqual(mod.key_color("-"), "")
+        colors = {mod.key_color(name) for name in ("a", "b", "c", "d", "e", "f")}
+        self.assertGreater(len(colors), 1)
+
     def test_print_logs_once_renders_requested_columns(self) -> None:
         mod = load_module()
         out = io.StringIO()
@@ -448,6 +484,10 @@ class Sub2APILogsTests(unittest.TestCase):
         self.assertIn("5.7K", text)
         self.assertIn("$0.012", text)
         self.assertIn("5321ms", text)
+        self.assertIn("first", text)
+        self.assertIn("tok/s", text)
+        self.assertIn("800ms", text)
+        self.assertIn("75.2/s", text)
         self.assertIn("total 2.3K records", text)
 
     def test_default_logs_token_reads_config_file(self) -> None: