fix(auth): prevent ERR_SERVER_NOT_RUNNING on magic-link login; harden stopServer

[facundo-herodevs]

Closes https://github.com/neverendingsupport/data-and-integrations/issues/549

Summary

This PR fixes a local CLI auth regression where auth login could throw:

Error [ERR_SERVER_NOT_RUNNING]: Server is not running.

even though authentication completed successfully (most visible in magic-link flows).

Root Cause

The OAuth callback server shutdown (server.close) was not resilient to re-entrant/racing shutdown paths:

• callback success path triggered shutdown
• error path could also trigger shutdown shortly after

When close() was called on an already-closed server, Node returned ERR_SERVER_NOT_RUNNING.
Because shutdown was not fully guarded, this surfaced as an exception during CLI login.

Changes

• Added stopServerPromise to deduplicate concurrent shutdown attempts.
• Refactored stopServer() to be idempotent:
  • capture server reference once
  • clear this.server early
  • return existing in-flight shutdown promise when present
• Treated ERR_SERVER_NOT_RUNNING as benign during cleanup.
• Kept user-facing warning generic (Failed to stop local OAuth callback server.) and moved detailed error info to debug logs.
• Updated non-awaited shutdown calls to explicit fire-and-forget (void this.stopServer()) where applicable.

Why This Approach

• Prevents false-negative CLI failures during successful auth flows.
• Makes shutdown logic deterministic and safe under event races.
• Keeps CLI output clean for users while preserving debuggability for maintainers.