refactor: comments and notifications doctype

[RitvikSardana]

Issue

Migrating helpdesk specific doctypes to their core equivalents.

• HD Ticket Comment to Comment
• HD Ticket Activity to Comment (Info), only tag, auto close and split rows
• HD Notification to Notification Log
• HD Comment Reaction reparented to Comment
• File attachments repointed to Comment

Patches

• setup_core_comments_schema: custom fields, notification types, permissions
• migrate_ticket_comments_to_comment: name preserving insert select
• migrate_ticket_activities_to_info_comments: rows with no Version row of their own
• repoint_comment_reactions_and_files: reactions and attachments
• migrate_hd_notifications_to_notification_log: notifications, with reopen rows retyped
• relabel_comment_search_index: relabels stale FTS rows in place, falls back to a rebuild on failure

Notes

• Comment names are preserved so existing mention links keep resolving
• Every patch is chunked (50k) and idempotent, so an interrupted bench migrate resumes
• Hash collisions are re detected on immutable fields and re inserted under a fresh name
• Old patch entries are untouched, this only appends

Inside the patch: one HD Ticket Comment, one fate

What copy_chunk and resolve_name_collisions decide for each row on its way to core Comment. Every path is safe to reach twice, which is the whole idempotency story.

flowchart TD
    S["Next batch: 50,000 HD Ticket Comments<br/><i>name &gt; lower AND name ≤ upper</i>"] --> Q1{"Same name already exists<br/>as a core Comment?<br/><i>NOT EXISTS check</i>"}
    Q1 -- "no" --> INS["INSERT as core Comment,<br/>name + creation preserved"]
    INS --> COMMIT["Chunk commits,<br/>resumable checkpoint"]
    Q1 -- "yes, skipped by copy_chunk" --> Q2{"That core Comment has my creation<br/>and points at HD Ticket?"}
    Q2 -- "yes" --> MINE["It's my own migrated copy<br/>(re-run / resumed run), do nothing"]
    Q2 -- "no, a stranger owns my name" --> Q3{"Core Comment copy of me exists<br/>under ANY name?<br/><i>creation + owner fingerprint</i>"}
    Q3 -- "yes" --> FIXED["Prior run already resolved me,<br/>do nothing"]
    Q3 -- "no" --> FRESH["INSERT core Comment under fresh hash,<br/>relink reactions · files · notifications,<br/>log the rename"]

    style Q1 stroke:#9a6b1f,stroke-width:2px
    style Q2 stroke:#9a6b1f,stroke-width:2px
    style Q3 stroke:#9a6b1f,stroke-width:2px
    style INS stroke:#2e7d4f,stroke-width:2px
    style FRESH stroke:#2e7d4f,stroke-width:2px
    style MINE stroke-dasharray: 5 4
    style FIXED stroke-dasharray: 5 4

Loading

Crash anywhere? Committed chunks stay, the interrupted chunk rolled back atomically. Re-running bench migrate replays the same tree: finished rows land on a dashed no-op, unfinished rows take the green path. Same statement, same answer, no duplicates and no manual repair.

depends-on: frappe/frappe#41924