feat(names): filter display names on a canonical form instead of a hardcoded list

[doopey655]

The display name filter is an inline List<string> in WebSocketController matched with
Contains(), so Нitler (Cyrillic Н), ＨＩＴＬＥＲ, h.i.t.l.e.r and hiiitler all pass, while
ibra rejects Ibrahim.

Names are now matched on a canonical form — NFKD, Unicode confusables folded to ASCII, leet fold,
repeats collapsed, non-alphanumerics stripped. Rule patterns get the same fold, so those spellings
reduce to one pattern.

Rules live in data/namefilter_rules.txt, one per line as action, match, pattern, category
(allow/block/review/shadow × skeleton/word/exact). It ships with the same 10 rules the
hardcoded list enforced. Edit the file and run !namefilter reload; a rule's id is its line number,
so !namefilter list and the scan report point at the line to change. A malformed line is logged
with its line number and skipped, and a file that cannot be read at all leaves the rules already
loaded in force rather than turning the filter off.

Also: users.displayname_skeleton for homoglyph impersonation, a reject log instead of telling the
user which pattern matched, rate limiting, and !namefilter rescan for names that predate a rule.

Run this before starting the service (also in the diff as
Database_Structure/migrations/2026-08-24-name-filter.sql). MySQL 8; the ALTER fails if it has
already been applied, which is the intended signal.

Migration

CREATE TABLE IF NOT EXISTS `name_filter_rejects` (
  `id` bigint(20) NOT NULL AUTO_INCREMENT,
  `user_id` bigint(20) NOT NULL,
  `attempted_name` varchar(64) NOT NULL DEFAULT '',
  `skeleton` varchar(64) NOT NULL DEFAULT '',
  `rule_id` int(11) NOT NULL DEFAULT -1,
  `action` tinyint(4) NOT NULL DEFAULT 1,
  `source` tinyint(4) NOT NULL DEFAULT 0,
  `created` datetime NOT NULL DEFAULT current_timestamp(),
  PRIMARY KEY (`id`),
  KEY `idx_user` (`user_id`),
  KEY `idx_created` (`created`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_general_ci;

ALTER TABLE `users`
  ADD COLUMN `displayname_skeleton` varchar(32) DEFAULT NULL AFTER `displayname`,
  ADD INDEX `idx_displayname_skeleton` (`displayname_skeleton`);

[Skyaero42]

I think a resource file would work much better here than a database.

[doopey655]

Agreed, and done — the rules are now data/namefilter_rules.txt rather than a table. Nothing joined against it in SQL, the rules are cached in memory after startup anyway, and data/ is already where this repo keeps its config. It ships in the publish output next to motd.txt, so it can be edited on the host and picked up with !namefilter reload.

One consequence worth naming: nothing in the service writes to data/, so there are no bot commands that add or remove rules — a rule change is an edit to the file. !namefilter decisions now prints the allow lines to paste in instead of writing them itself. Rule ids are line numbers in the file, so list and the scan report point at the line to edit.

The rules table and its seed are out of the migration; name_filter_rejects and users.displayname_skeleton are still there.