feat(model-eval): expose per-attempt benchmark metrics (#3477)

* feat(model-eval): expose per-attempt benchmark metrics

* refactor(model-eval): omit cost coverage metric

* chore(db): add model eval attempt metric migration

* refactor(model-eval): narrow promotion refresh to named repulls

Author: lambertjosh

packages/db/src/migrations/0144_happy_captain_america.sql

@@ -0,0 +1,5 @@
+ALTER TABLE "model_eval_ingestions" ADD COLUMN "n_attempts" integer;--> statement-breakpoint
+ALTER TABLE "model_eval_ingestions" ADD COLUMN "total_cost_microdollars" bigint;--> statement-breakpoint
+ALTER TABLE "model_eval_ingestions" ADD COLUMN "total_input_tokens" bigint;--> statement-breakpoint
+ALTER TABLE "model_eval_ingestions" ADD COLUMN "total_output_tokens" bigint;--> statement-breakpoint
+ALTER TABLE "model_eval_ingestions" ADD COLUMN "total_cache_read_tokens" bigint;