refactor(automod): remove regex classifier — LLM is the sole judge

Delete fastClassifier.ts (manual regex patterns for phone/email/IP/crypto/
spam/toxicity) and simpleFallback.ts. These hardcoded patterns were the
source of false positives (Discord emoji snowflakes matched phone_number,
URL digits matched phone, etc.) and produced heuristic verdicts whenever
the LLM failed.

New flow: Message → LLM (with conversation context, media evidence, user
reputation) → verdict. On LLM failure the message is marked 'error' and
retried by the recovery worker — no heuristic verdicts, ever.

Discord markdown tokens (custom emoji/mentions/timestamps) are normalized
to readable placeholders ([emoji:name], @user, @role, #channel, [time])
before reaching the LLM via discordTokens.ts.

services/discord-gateway/src/modules/ai-moderation/conversationContext.ts

@@ -2,6 +2,7 @@ import { createChildLogger } from "@/shared/logger/index";
 import { encoding_for_model as encodingForModel } from "tiktoken";
 import { formatMediaEvidenceForPrompt } from "../message-capture/messageMetadata.js";
 import type { MessageRecord } from "../message-capture/types.js";
+import { sanitizeDiscordTokens } from "./discordTokens.js";
 
 const logger = createChildLogger("conversationContext");
 
@@ -67,7 +68,9 @@ export function formatMessageForPrompt(
   msg: MessageRecord,
   label: "context" | "target",
 ): string {
-  const content = msg.edited_content ?? msg.content;
+  const content = sanitizeDiscordTokens(
+    msg.edited_content ?? msg.content,
+  );
   const timestamp = formatTimestamp(msg.created_at);
   const mediaEvidence = formatMediaEvidenceForPrompt(msg.metadata);
   const mediaSuffix = mediaEvidence ? ` ${mediaEvidence}` : "";