Add service layer orchestrating comment queries and thread assembly:

comment_service.go (285 lines):
- GetComments: Main query method with validation and pagination
- buildThreadViews: Recursively constructs comment trees
- Iterative loading strategy (loads 5 replies per level)
- Respects depth limit (default 10, max 100)
- Sets HasMore flag for pagination hints
- buildCommentView: Converts entities to API views
- Hydrates author from CommenterHandle
- Builds stats (upvotes, downvotes, score, replyCount)
- Creates post/parent references with CIDs
- Stub viewer state (Phase 2B)
- validateGetCommentsRequest: Input validation with defaults

view_models.go (150 lines):
- CommentView: Complete comment with author, stats, viewer state
- ThreadViewComment: Recursive wrapper for nested replies
- Supporting types matching lexicon definitions
- Follows existing patterns from posts.AuthorView

Changes to existing files:
- comment.go: Add CommenterHandle field (hydrated at query time)
- errors.go: Add IsValidationError helper for handler error mapping

Design decisions:
- Empty slices instead of nil (JSON serialization)
- Iterative loading prevents N+1 query explosion
- Soft-deleted comments filtered out
- Post/user integration stubbed (Phase 2C)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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

Implement database layer for comment queries with Lemmy hot ranking:

New repository methods:
- ListByParentWithHotRank: Query with hot/top/new sorting + pagination
- Hot: log(greatest(2, score + 2)) / power(time_decay, 1.8)
- Top: Score-based with optional timeframe filter
- New: Chronological ordering
- Cursor-based pagination with composite keys

- GetByURIsBatch: Batch fetch comments by URIs (prevents N+1 queries)
- GetVoteStateForComments: Fetch viewer votes (Phase 2B ready)

Key features:
- Hydrates author handle via JOIN with users table
- Supports timeframe filters (hour/day/week/month/year/all)
- Encodes cursors with hot_rank|score|created_at|uri
- All queries use parameterized arguments (SQL injection safe)

Formula prevents brigading:
- greatest(2, score + 2) ensures log never goes negative
- Heavily downvoted comments bounded at log(2)
- Power of 1.8 for faster decay than posts (1.5)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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

Add lexicon definitions for comment query API following Bluesky patterns:

- social.coves.community.comment.defs: Shared view definitions
- commentView: Base view for single comment with stats and viewer state
- threadViewComment: Recursive wrapper for threaded replies
- Supporting types: commentStats, commentViewerState, commentRef, etc.

- social.coves.community.comment.getComments: Query endpoint
- Parameters: post (required), sort, depth, limit, cursor, timeframe
- Returns threaded comments with nested replies up to depth limit
- Supports hot/top/new sorting with Lemmy-style hot ranking

Follows atproto best practices:
- Composition pattern (threadView wraps baseView)
- Union types for error states (notFound, blocked)
- Open unions for future extensibility
- Strong references with CID version pinning

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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

Document P1 issue discovered during comment system implementation:
when comments arrive before their parent post (cross-repo Jetstream ordering),
the post's comment_count is never reconciled.

Issue details:
- Comment consumer updates post counts when processing events
- If comment arrives BEFORE post is indexed, update returns 0 rows
- When post consumer later indexes the post, it sets comment_count = 0
- NO reconciliation logic to count pre-existing comments

Solution: Post consumer must implement same reconciliation pattern as
comment consumer (COUNT subquery after insert).

Related: Comment reply_count reconciliation was fixed in comment system
implementation (2025-11-04).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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

Initialize comment repository and Jetstream consumer at server startup.
Consumer runs in background goroutine, indexing comment events from
atProto firehose to PostgreSQL AppView.

Consumer lifecycle:
- Start on server init
- Graceful shutdown on SIGINT/SIGTERM
- Automatic reconnection on connection loss

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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

Add 20 integration tests covering all comment indexing scenarios:

Core operations:
- Create comment (normal, idempotent, out-of-order)
- Update comment (content, metadata)
- Delete comment (soft delete)

Threading:
- Root/parent references
- Reply count updates
- Thread hierarchy queries

Security:
- Invalid DID rejection
- Content length limits
- Malformed AT-URI rejection
- Threading immutability (reject mutation attempts)

Out-of-order handling:
- Child arrives before parent (count reconciliation)
- Multiple children before parent

Resurrection:
- Recreate deleted comment (same parent)
- Recreate deleted comment (different parent - tests threading ref updates)

Repository queries:
- ListByRoot, ListByParent, ListByCommenter
- Soft delete filtering

All tests verify both database state and denormalized counts.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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