feat: Store Bot Comment Embeddings from Closed PRs

[RulaKhaled]

• Add tests

This PR adds functionality to collect and store bot comment patterns when PRs are closed. We analyze user reactions to identify good/bad patterns and store them with embeddings for future similarity matching.

Implementation Details

1. PR Closed Handler

class PrClosedStep(CodegenStep):
    """Processes closed PRs to collect comment patterns"""

• Triggers when a PR is closed
• Fetches all review comments from the PR
• Filters for bot comments only (codecov-ai-reviewer[bot])

2. Comment Analysis

• Counts 👍 and 👎 reactions on each comment
• Determines if it's a good pattern (more 👍) or bad pattern (more 👎)

3. Embedding Generation

# Generate embeddings for similarity matching
model = GoogleProviderEmbeddings.model(
    "text-embedding-005", 
    task_type="CODE_RETRIEVAL_QUERY"
)
embedding = model.encode(comment.body)[0]  # 768-dimensional vector

4. Database Storage

class DbReviewCommentEmbedding(Base):
    """Store PR review comments with their embeddings for pattern matching per organization"""
    __tablename__ = "review_comments_embedding"

• Stores comment text, metadata, and embeddings
• Uses pgvector for efficient similarity search
• Handles duplicates with on_conflict_do_nothing

Example Flow

1. PR Beta readiness #123 is closed
2. System finds bot comment
3. Comment has 3 👍 and 1 👎 → marked as good pattern
4. Generates embedding for similarity matching
5. Stores in DB with metadata (PR link, file location, etc.)

[rohitvinnakota-codecov]

I think we can just use PR review credentials. That way we don't have to add additional env values