Fix thread-safety issues in native bindings replacing Nan::Persistent with v8::Eternal

[rmnilin]

This PR addresses thread-safety issues in the native bindings of the library. It replaces Nan::Persistent with v8::Eternal for constructor storage in cipher and decipher classes.

The previous implementation using Nan::Persistent was not thread-safe, leading to crashes and undefined behavior when these classes were accessed from multiple threads simultaneously. This was particularly problematic when ssh2 was used in multi-threaded environments, such as when imported in worker threads.

Key changes:

1. Modified the constructor method to return a v8::Eternal instead of a Nan::Persistent.
2. Updated the init function to use the Set method of the Eternal handle.

These changes ensure thread-safe storage of constructor handles without altering the intended functionality of the modules.

This PR addresses issue #1393 and resolves the reported crashes and segmentation faults related to multi-threaded use of the ssh2 library.

[mscdex]

I don't understand how using a v8::Eternal fixes a thread safety issue. AFAIK the only real difference between v8::Eternal and v8::Persistent is the lifetime of the value. v8::Eternal will always last the lifetime of the isolate whereas v8::Persistent will not necessarily.

[mscdex]

Is the underlying reason for the worker-related issues that constructor() returns a static v8::Persistent?

[rmnilin]

Thank you for the feedback!

Yes, the underlying issue with worker threads is related to the use of a static Nan::Persistent. This static handle could lead to race conditions and undefined behavior in a multi-threaded environment. By switching to v8::Eternal, we ensure that the constructor handle remains stable and safe across multiple threads, as v8::Eternal ties the handle's lifetime to the isolate, avoiding the problems associated with the static storage duration of Nan::Persistent.

While the primary distinction between Nan::Persistent and v8::Eternal is indeed the lifetime, v8::Eternal provides a more reliable guarantee for thread safety in multi-threaded environments, addressing the described issues.

I've also created rmnilin/ssh2-1393-repro to test both the current implementation and this fix.