fix: guard against NaN in confidenceScore and fix Query 8 comment (CM-1137)

- Replace `??` with explicit isFinite check so NaN from edit-distance
  calc (identityLength=0, smallestEditDistance=null) does not propagate
  as confidenceScore — falls back to HIGH_CONFIDENCE_SCORE instead
- Correct Query 8 comment: noreplyEmailUsernameMatches is populated only
  inside the verified-identity branch, not for unverified identities

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: skwowet

services/apps/merge_suggestions_worker/src/activities/memberMergeSuggestions.ts

@@ -217,7 +217,7 @@ export async function getMemberMergeSuggestions(
       }),
     },
     {
-      // Query 8: Noreply/private email -> username (verified or unverified)
+      // Query 8: Noreply email -> platform username (verified identities only)
       matches: uniqBy(noreplyEmailUsernameMatches, (m) => `${m.platform}:${m.value}`),
       builder: ({ value, platform }) => ({
         bool: {

services/apps/merge_suggestions_worker/src/memberSimilarityCalculator.ts

@@ -467,7 +467,10 @@ class MemberSimilarityCalculator {
     // displayName equality (startingScore omitted) is itself a high-confidence signal — don't
     // gate on metadata. Edit-distance callers pass a score and must have at least one signal.
     let isHighConfidence = startingScore === undefined
-    let confidenceScore = startingScore ?? this.HIGH_CONFIDENCE_SCORE
+    let confidenceScore =
+      startingScore != null && Number.isFinite(startingScore)
+        ? startingScore
+        : this.HIGH_CONFIDENCE_SCORE
 
     const bumpFactor = (1 - confidenceScore) / 5