Cache exact Frontier EP lane predictions

runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_rank.py

@@ -631,6 +631,7 @@ def run_frontier(args: argparse.Namespace) -> Path:
             "quantization": "block_fp8_w8a8_dynamic_with_bf16_output_accumulation",
             "execution_profiles": "community_vllm_0.10.2_serving_entrypoints",
             "moe_ep_prefill": "global_routing_then_slowest_local_ep_lane",
+            "moe_ep_critical_lane_cache": "exact_(total_routed_tokens,layer_id)_memoization",
             "kv_blocks": "measured_community_vllm_and_fixed_per_topology",
             "cpu_overhead_modeling": "disabled_no_native_community_vllm_records",
             "end_to_end_action_specific_calibration": False,

runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_exact_cache.patch

@@ -0,0 +1,17 @@
+diff --git a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py
+--- a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py
++++ b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py
+@@ -1,4 +1,5 @@
+ from typing import Any, Dict, List, Optional, TYPE_CHECKING, Union
++from functools import lru_cache
+ import os
+ 
+ import numpy as np
+@@ -384,6 +385,7 @@
+             allocation_ratios=allocation_ratios,
+         )
+ 
++    @lru_cache(maxsize=None)
+     def _get_critical_prefill_ep_lane_tokens(
+         self,
+         total_routed_tokens: int,