feat: add Docker entrypoints, LLM providers, pipeline hardening, workbench pages

Phase 9 — four parallel workstreams:

- Stream A: 14 Docker entrypoints for containerized deployment
- Stream B: Pipeline hardening — robust JSON parsing, LLM retry logic,
  consumer negative-ack, FalkorDB test import fix
- Stream C: Azure OpenAI, OpenAI-compatible, and Mistral LLM providers
- Stream D: Workbench Prompts, Token Cost, Knowledge Cores pages +
  Settings feature switches

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

ts/packages/base/src/messaging/consumer.ts

@@ -73,8 +73,9 @@ export class Consumer<T> {
 
   private async consumeLoop(flow: FlowContext): Promise<void> {
     while (this.running) {
+      let msg: Message<T> | null = null;
       try {
-        const msg = await this.backend!.receive(2000);
+        msg = await this.backend!.receive(2000);
         if (!msg) continue;
 
         await this.handleWithRetry(msg, flow);
@@ -82,7 +83,13 @@ export class Consumer<T> {
       } catch (err) {
         if (!this.running) break;
         console.error("[Consumer] Error in consume loop:", err);
-        // Brief pause before retry
+        if (msg) {
+          try {
+            await this.backend!.negativeAcknowledge(msg);
+          } catch (nakErr) {
+            console.error("[Consumer] Failed to nak message:", nakErr);
+          }
+        }
         await sleep(1000);
       }
     }