Give the AI subsystem one profile table and one collection, read everywhere

Plugin/unraid/Tools/api_cache_writer.php

@@ -91,19 +91,23 @@ $t = microtime(true);
 // Call vv_api_data() once — result is static-cached for the rest of this process.
 vv_api_data();
 
-// Must stay in step with api/monitor.php's own block. This file is what the Monitor tab
-// normally reads — the endpoint only assembles a payload on a cache miss — so a key added there
-// and not here leaves the card that consumes it loading forever on every ordinary page load,
-// and working on the one request that happens to miss the cache.
+// ── AI ────────────────────────────────────────────────────────────────────────
+// One collection, two consumers. vv_ai_stats() is the expensive part of the AI subsystem —
+// roughly a second, most of it waiting on Ollama and nvidia-smi — and it is written to its own
+// cache here so the AI tab's banner, the Scheduler dock and the Monitor row all read the same
+// numbers from the same moment instead of each paying for their own.
+//
+// The monitor block is derived from that same array rather than collected again. Must stay in
+// step with api/monitor.php's own block: this file is what the Monitor tab normally reads, since
+// the endpoint only assembles a payload on a cache miss, so a key added there and not here
+// leaves the card that consumes it loading forever on every ordinary page load and working only
+// on the one request that happens to miss.
 $_vv_ai = null;
 if (vv_ai_ui_on()) {
     require_once $_base . '/include/ai.php';
-    $_vv_ai = [
-        'model'   => vv_ai_config()['model'],
-        'runtime' => vv_ai_runtime_stats(),
-        'index'   => vv_ai_index_stats(),
-        'tokens'  => vv_ai_token_stats()['today'] ?? null,
-    ];
+    $_vv_ai_stats = vv_ai_stats();
+    vv_cache_write('ai', $_vv_ai_stats);
+    $_vv_ai = vv_ai_monitor_block($_vv_ai_stats);
 }
 
 $monitor = [