fix(lib-storage): reject done() with TimeoutError when CompleteMultipartUpload hangs

[Zelys-DFKH]

Issue

#7729

Description

Upload.done() hangs forever when the TCP connection carrying the CompleteMultipartUpload response is silently dropped by a network intermediary (firewall, DLP appliance, transparent proxy) while S3 assembles a large object server-side. This is common on uploads over ~15 minutes: all parts upload successfully, the complete call goes out, but the HTTP response never arrives because the proxy killed the idle connection. With no timeout on that client.send() call, the promise stalls indefinitely.

This adds a completeMultipartUploadTimeout option (milliseconds) to Upload. When set, the CompleteMultipartUpload send races against a setTimeout; if the timer fires first, done() rejects with a TimeoutError. Without the option, behavior is identical to before: no timeout imposed.

The workaround from #7729 (configure keepAlive: true on the HTTP agent) still works. This option is complementary for callers who can't control the network layer or who want a hard deadline.

Credit: @paffoobar's root-cause analysis in #7729.

This change was generated with AI assistance and reviewed by me.

Testing

Added two unit tests to Upload.spec.ts (54/54 passing):

• Timeout set + CompleteMultipartUpload hangs indefinitely: rejects with TimeoutError after the configured milliseconds
• No timeout set: resolves normally regardless of how long CompleteMultipartUpload takes (backward-compatible path verified)

Checklist

• It's not a feature.
• I didn't write any E2E tests.
• I didn't add any public functions.
• No streams here.