fix(eventcounter): deduplicate events by ID in BigQuery count queries

BigQuery has no primary keys, so an at-least-once Pub/Sub redelivery
after a successful append stores the same event twice. The MySQL and
Postgres event tables reject duplicates via PRIMARY KEY (id), but the
BigQuery count queries trusted row uniqueness:

- evaluation_count.sql used COUNT(id) for the event total, so a
  duplicated evaluation event inflated evaluationTotal forever.
- goal_count.sql aggregated COUNT(id) and SUM(value) per user directly
  over the raw table, so duplicates inflated goalTotal and the per-user
  value sums that feed the Bayesian value-metric analysis.

User counts (COUNT(DISTINCT user_id)) and therefore the CVR analysis
were already duplicate-tolerant.

Deduplicate by event ID: COUNT(DISTINCT id) for the evaluation total,
and a DISTINCT-by-ID CTE in the goal query before any aggregation. Each
legitimate event has its own unique ID, so repeat evaluations/goals by
the same user are still counted. This also retroactively corrects
duplicates already stored in the tables.

This is also what made TestGrpcExperimentResult hang in the dev cluster:
the test waits for exact event counts, and a single duplicated event
kept the count above the expected total forever.

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: cre8ivejp

pkg/eventcounter/storage/v2/dwh_database/bigquery/sql/evaluation_count.sql

@@ -1,7 +1,11 @@
 SELECT
     variation_id as variationID,
     COUNT(DISTINCT user_id) as evaluationUser,
-    COUNT(id) as evaluationTotal
+    -- DISTINCT by event ID: BigQuery has no primary keys, so an at-least-once
+    -- Pub/Sub redelivery can append the same event twice. Each legitimate
+    -- event has its own unique ID, so this still counts repeat evaluations
+    -- by the same user.
+    COUNT(DISTINCT id) as evaluationTotal
 FROM
     `%s`
 WHERE

pkg/eventcounter/storage/v2/dwh_database/bigquery/sql/goal_count.sql

@@ -1,9 +1,14 @@
-WITH grouped_by_user_evaluation AS (
-    SELECT
+WITH deduped_events AS (
+    -- BigQuery has no primary keys, so an at-least-once Pub/Sub redelivery
+    -- can append the same event twice. Deduplicate by event ID before any
+    -- aggregation, otherwise duplicates inflate event_count and value_sum.
+    -- Duplicates are identical rows, so DISTINCT over the used columns is
+    -- equivalent to picking one row per ID.
+    SELECT DISTINCT
+        id,
         user_id,
         variation_id,
-        COUNT(id) as event_count,
-        IFNULL(SUM(value), 0) as value_sum
+        value
     FROM
         `%s`
     WHERE
@@ -12,9 +17,18 @@ WITH grouped_by_user_evaluation AS (
     AND goal_id = @goalID
     AND feature_id = @featureID
     AND feature_version = @featureVersion
-GROUP BY
-    user_id,
-    variation_id
+),
+grouped_by_user_evaluation AS (
+    SELECT
+        user_id,
+        variation_id,
+        COUNT(id) as event_count,
+        IFNULL(SUM(value), 0) as value_sum
+    FROM
+        deduped_events
+    GROUP BY
+        user_id,
+        variation_id
 ),
 cap_level AS (
     -- Winsorization threshold: the configurable percentile