Summary

• Jira often emits multiple changelog items with the same field in one history entry (add/remove of Fix Version, Component, labels, etc.). _tool_jira_issue_changelog_items was keyed by (connection_id, changelog_id, field), so the extractor silently kept only one row.
• Add a per-field item_index to the primary key. Indexes are assigned from item content (from/to values and strings), not JSON array order, so re-extract of the same changelog stays stable if Jira reshuffles items.
• Domain issue_changelogs.id includes item_index. The migration rebuilds the PK, resets Jira extract/convert subtask state so the next pipeline re-extracts from existing raw JSON, and drops old domain rows that used the previous id format.

Fixes #9057