asim/state: use epsilon in cumulative weights checking

Previously, we asserted that cumulative weights in the weighted random
distribution must sum to exactly 1. Due to floating point precision limitations,
this is not always achievable and can lead to false failures like total
cumulative weights for all stores should sum up to 1 but got 1.00. This commit
relaxes the check to allow a small epsilon tolerance in the comparison to avoid
such errors.

Author: wenyihu6

pkg/kv/kvserver/asim/state/new_state.go

@@ -102,7 +102,8 @@ func newWeighted(weightedStores []float64) weighted {
 		prefixSumWeight += item
 		cumulativeWeights[i] = prefixSumWeight
 	}
-	if cumulativeWeights[len(weightedStores)-1] != float64(1) {
+	const epsilon = 1e-10
+	if math.Abs(cumulativeWeights[len(weightedStores)-1]-float64(1)) > epsilon {
 		panic(fmt.Sprintf("total cumulative weights for all stores should sum up to 1 but got %.2f\n",
 			cumulativeWeights[len(weightedStores)-1]))
 	}