feat: wire costs[] from modelUsage into eval results

Extract per-model token usage from resultLine.modelUsage (including
cache tokens and exact API cost), flow CostEntry[] through EvalCollector,
aggregate in finalize(). Extend CostEntry with cache_read_input_tokens,
cache_creation_input_tokens, cost_usd. computeCosts() prefers exact
cost_usd over MODEL_PRICING when available (~4x more accurate with
prompt caching).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

test/skill-e2e.test.ts

@@ -41,6 +41,7 @@ function recordE2E(name: string, suite: string, result: SkillTestResult, extra?:
     exit_reason: result.exitReason,
     timeout_at_turn: result.exitReason === 'timeout' ? result.costEstimate.turnsUsed : undefined,
     last_tool_call: lastTool,
+    costs: result.costs,
     ...extra,
   });
 }