fix: review-agent dedup bot review bodies with inline comments (#78134)

The comment analyzer flags review bodies and review threads
independently. When an approved bot (e.g. CodeRabbit) submits a review
with both a body and inline comments, both get flagged as separate
needs_attention items. Claude addresses each independently, producing
duplicate responses.

Observed on openshift/hypershift#8280: CodeRabbit review #4146142149
had a body ("Actionable comments posted: 1") and one inline comment
about trust bundle sync stalling. The analyzer returned both as
separate items. Claude posted an inline reply to the thread at
11:04:10Z, then two issue comments at 11:04:20Z and 11:06:19Z —
both "Re: #pullrequestreview-4146142149" with overlapping content.

Root cause: analyze_review_bodies uses per-comment reply matching
(checking if any bot comment references the review ID/URL). Since no
bot comment had yet referenced review 4146142149, it was flagged.
Meanwhile, analyze_review_threads independently flagged the same
review's inline thread. Both items entered the needs_attention list.

Fix: add comments { totalCount } to the reviews GraphQL query and
skip approved bot review bodies when inline comments exist. Bot
review bodies are machine-generated summaries of their inline
comments (already handled by analyze_review_threads). Human review
bodies are always kept since they often contain distinct feedback.

Verified against PR #8280:
- CodeRabbit review 4146142149: SKIPPED (approved_bot, inline_count=1)
- CodeRabbit review 4138831511: SKIPPED (approved_bot, inline_count=1)
- csrwng review 4144456157: KEPT (human, inline_count=2)

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: bryan-cox

ci-operator/step-registry/hypershift/review-agent/process/hypershift-review-agent-process-commands.sh

@@ -417,6 +417,7 @@ def analyze_review_bodies(pr_number: int) -> list[dict]:
               body
               state
               submittedAt
+              comments(first: 0) { totalCount }
             }
           }
         }
@@ -463,6 +464,15 @@ def analyze_review_bodies(pr_number: int) -> list[dict]:
         if not is_authorized_author(author):
             continue
 
+        # Skip approved bot review bodies when inline comments exist.
+        # Bot review bodies (e.g. CodeRabbit's "Actionable comments posted: N")
+        # are machine-generated summaries of their inline comments. Those inline
+        # comments are already handled by analyze_review_threads, so flagging the
+        # body too causes Claude to address the same feedback twice.
+        inline_count = review.get("comments", {}).get("totalCount", 0)
+        if is_approved_bot(author) and inline_count > 0:
+            continue
+
         review_id = review["id"]
         review_url = review.get("url", "")