Fix: Ensure all 13 metric columns always present in datasets

CRITICAL FIX: Resolves missing CO2 and power cost columns in metrics dataset.

Root Cause:
- Metrics with empty dataPoints arrays were skipped during flattening
- CO2 and power cost are cumulative and start at 0
- If first metric batch exported before accumulation, columns missing
- Result: Datasets had only 11 columns instead of expected 13

Solution:
- Initialize ALL 7 metric columns with default 0.0 values before processing
- Metrics with data points overwrite defaults with actual values
- Metrics without data points keep default 0.0
- Ensures consistent 13-column schema in every dataset

Impact:
✅ Metrics dataset always has complete schema
✅ Leaderboard aggregation never fails due to missing columns
✅ TraceMind UI can rely on consistent data structure
✅ CO2 and power cost start at 0.0, increase during evaluation

Files Modified:
- smoltrace/utils.py: Initialize columns in flatten_metrics_for_hf()
- changelog.md: Document fix

Testing:
- Verified with 2-second test (only 5 metrics, but 13 columns created)
- All columns present with appropriate values

Author: kshitijthakkar

.claude/settings.local.json

@@ -0,0 +1,10 @@
+{
+  "permissions": {
+    "allow": [
+      "Read(//h/C_Documents/Documents/LLM-Code/llm_engineering/Projects/TraceMind_planning/.venv/Lib/site-packages/genai_otel/instrumentors/**)",
+      "Read(//h/C_Documents/Documents/LLM-Code/llm_engineering/Projects/TraceMind_planning/.venv/Lib/site-packages/genai_otel/**)"
+    ],
+    "deny": [],
+    "ask": []
+  }
+}