fix: format GNN training metrics timestamp with RFC3339 timezone

- Julia script now outputs UTC timestamps as RFC3339 with Z suffix
- Enables Rust parser DateTime::parse_from_rfc3339() to succeed
- gnn_training_integration_test now passes (last_trained field populated)

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

Author: hyperpolymath

models/training_metrics.json

@@ -0,0 +1 @@
+{"status":"completed","training_data_size":10011,"MRR":0.85,"model_location":"models/neural/gnn_ranker","timestamp":"2026-04-25T15:26:45.716Z","nDCG":0.85,"epochs_trained":50}

src/julia/train_and_evaluate.jl

@@ -78,7 +78,7 @@ function train_and_evaluate()
 
     # Create metrics structure
     metrics = Dict(
-        "timestamp" => string(now()),
+        "timestamp" => Dates.format(now(Dates.UTC), "yyyy-mm-ddTHH:MM:SS.sssZ"),
         "training_data_size" => samples,
         "epochs_trained" => epoch,
         "nDCG" => round(ndcg_score, digits=4),