Support custom router for producer

[shibd]

Motivation

To support a custom router in the Node.js client, the most direct approach would be to call the user-provided router function from within the C++ callback. However, because the C++ client's sendAsync method has a synchronous component for partition selection before it becomes fully asynchronous, this implementation causes a deadlock between the Node.js main thread and the C++ callback thread. This is due to the single-threaded nature of the Node.js event loop.

https://github.com/apache/pulsar-client-cpp/blob/90ea3695b80660f837785d98e96047d90de3f64f/lib/PartitionedProducerImpl.cc#L209-L220

You can understand by reviewing the PR and running the associated tests.

Modifications

This PR achieves the goal by calling the user's router function at the JS layer and then passing the resulting partition index to the C++ layer via message properties.

1. The send method in Producer.js is wrapped. It now first calls the user's router function to asynchronously determine the target partition.
2. The resulting partition index is added to the message's properties under the special key: __partition__.
3. Inside the N-API implementation, a fixed, synchronous C++ function (internalCppMessageRouter) is now used, which simply reads this property and returns the partition index to the underlying C++ client.

Points for Discussion

This implementation introduces two points for discussion:

1. We need to maintain a cache for the number of partitions (numPartitions) in Producer.js, which is then passed to the user's router function. (Currently, this cache has a 60-second TTL. A potential future improvement is to expose a method from the C++ producer to get the partition count directly. This would allow us to remove the JS-side cache, as the C++ client internally handles partition metadata updates.)
2. This introduces a special property (__partition__) in the message properties, which could potentially be confusing for users. However, I believe this is not a major issue at the moment.

Verifying this change

• The Message Routing test case has been added to cover this change.

Documentation

• doc-required
  (Your PR needs to update docs and you will update later)

• doc-not-needed
  (Please explain why)

• doc
  (Your PR contains doc changes)

• doc-complete
  (Docs have been already added)

[shibd]

I will push a separate PR to fix failed CI.

[BewareMyPower]

I have a better idea for this implementation, block this PR first. Please wait for my PR,

[shibd]

Support on PR: #435