feat: add destroy() to MdkNode to fix 402 zombie node race condition (#30)

* feat: add destroy() method to MdkNode for explicit cleanup

On serverless platforms (Vercel/Lambda), MdkNode's inner Rust Node
and its tokio runtime survive after getInvoice() returns because V8
GC is non-deterministic. When the agent pays a 402 invoice instantly
(<1s), the webhook handler creates a second MdkNode for the same
wallet while the first is still alive. The zombie's reconnection
loop steals the LSP peer connection from the new node, preventing
the JIT channel from being established and causing "retries exhausted"
payment failures.

In normal checkout this doesn't happen because the human delay (5-30s)
gives GC time to collect the old node before the webhook fires.

destroy() wraps the inner Node in Option<Node>, allowing JS callers
to explicitly drop the Rust Node and its tokio runtime immediately
after invoice creation, eliminating the race condition.

* style: fix cargo fmt

Author: martinsaposnic

index.d.ts

@@ -71,6 +71,12 @@ export interface NodeChannel {
 }
 export declare class MdkNode {
   constructor(options: MdkNodeOptions)
+  /**
+   * Destroy the node, dropping the inner Rust Node and its tokio runtime immediately.
+   * This prevents zombie processes on serverless platforms where GC is non-deterministic.
+   * After calling destroy(), any further method calls on this node will panic.
+   */
+  destroy(): void
   getNodeId(): string
   start(): void
   stop(): void