[lhk0504]

Fixes #1645

Proposed Changes

• Resolve the distinct targets of a user's available comments and reuse SimpleObjectInfo.CheckVisibility to select the targets visible to the current viewer.
• Apply the visible target IDs to the comment query before pagination and counting. In the reported A/B/C example, page 1 now contains C with count: 1, and page 2 is empty with count: 1.
• Reuse the resolved object information when building the response. Existing comment-page callers remain unrestricted when the object-ID filter is nil, while an empty filter returns no comments.
• Add real SQLite regression tests covering pagination, repeated targets, user and comment-status isolation, empty and missing targets, question and answer ownership, moderator access, hidden parent questions, closed questions, and tag targets.

Validation

Passed locally:

• go test ./internal/service/comment ./internal/repo/repo_test -count=1 (SQLite)
• go build ./...
• go vet ./...
• golangci-lint run ./internal/service/comment/... ./internal/repo/comment/... ./internal/repo/repo_test/...
• git diff --check

The new service regression tests were also run against the old implementation and reproduced the incorrect counts and pagination.

MySQL/PostgreSQL tests were not run because the local Docker socket is inaccessible.

Trade-off

Each request checks all distinct comment targets for the profile user before querying the requested page. Request cost and the size of the SQL IN filter therefore grow with the number of distinct targets; large-scale performance has not been benchmarked.

Related: #1648 proposes another fix for the same issue.