fix(prompt-budget): count all startup skill metadata

adapters/prompt-budget-baseline.json

@@ -4,11 +4,15 @@
   "permanent_context_hard_limit_tokens": 6000,
   "metrics": {
     "permanent_context_tokens": {
-      "claude": 1481,
-      "codex": 1481
+      "claude": 4029,
+      "codex": 2603
+    },
+    "instruction_tokens": {
+      "claude": 1479,
+      "codex": 1479
     },
     "skill_metadata_tokens": {
-      "claude": 1124,
+      "claude": 2550,
       "codex": 1124
     },
     "rule_catalog_tokens": {

adapters/prompt-budget-baseline.tsv

@@ -1,7 +1,9 @@
 metric	client	baseline_tokens
-permanent_context_tokens	claude	1481
-permanent_context_tokens	codex	1481
-skill_metadata_tokens	claude	1124
+permanent_context_tokens	claude	4029
+permanent_context_tokens	codex	2603
+instruction_tokens	claude	1479
+instruction_tokens	codex	1479
+skill_metadata_tokens	claude	2550
 skill_metadata_tokens	codex	1124
 rule_catalog_tokens	claude	28036
 rule_catalog_tokens	codex	8058