fix: use a short-lived AbortSignal for fetch operations (#511)

Node's `undici` module appears to be adding listeners for the `"abort"` event on any passed AbortSignal but then not cleaning them up.

This casues many `MaxListenersExceededWarning`s to appear in the console and likely causes a memory leak.

To work around this, use a short-lived `AbortSignal` for each request that can be garbage collected along with the request itself.

See https://github.com/nodejs/node/issues/52635 for more information.

packages/block-brokers/src/trustless-gateway/trustless-gateway.ts

@@ -91,12 +91,20 @@ export class TrustlessGateway {
     }
 
     const blockId = this.#uniqueBlockId(cid)
+
+    // workaround for https://github.com/nodejs/node/issues/52635
+    const innerController = new AbortController()
+    const abortInnerSignal = (): void => {
+      innerController.abort()
+    }
+    signal?.addEventListener('abort', abortInnerSignal)
+
     try {
       let pendingResponse: Promise<Uint8Array> | undefined = this.#pendingResponses.get(blockId)
       if (pendingResponse == null) {
         this.#attempts++
         pendingResponse = fetch(gwUrl.toString(), {
-          signal,
+          signal: innerController.signal,
           headers: {
             Accept: 'application/vnd.ipld.raw'
           },
@@ -122,6 +130,7 @@ export class TrustlessGateway {
       this.#errors++
       throw new Error(`unable to fetch raw block for CID ${cid}`)
     } finally {
+      signal?.removeEventListener('abort', abortInnerSignal)
       this.#pendingResponses.delete(blockId)
     }
   }