perf(zql): optimize Join processParentNode closure and spread

[Karavil]

Note: This PR was generated by Claude (Anthropic) and has not been
reviewed by @Karavil. It is part of an upstream contribution effort from
the Goblins team.

Summary

Optimize Join#processParentNode by extracting the child stream closure to a method and skipping the object spread when parentNodeRelations is empty.

Motivation

#processParentNode is called for every parent row during join fetch and push operations. In a workload with 135 IVM pipelines, many of which involve joins with ~200 child rows per parent, this method is called tens of thousands of times per page render.

Two specific costs:

1. Closure allocation: The current code creates a new closure inside #processParentNode on every call. Extracting the logic to a class method (#fetchChildStream) means we create a lightweight arrow function that delegates to the method, rather than a closure capturing local constraint-building logic.
2. Object spread on empty: For leaf joins (the most common case in deep join pipelines), parentNodeRelations is an empty {}. The {...parentNodeRelations, [name]: stream} spread still allocates a new object and copies all (zero) properties. Checking emptiness with for...in and skipping the spread avoids this overhead.

Changes

• Extract inline closure to #fetchChildStream(parentNodeRow) private method
• Add fast path in #processParentNode: check if parentNodeRelations has any keys via for...in, skip the spread if empty

Expected Performance Impact

For leaf joins (bottom of a join chain), the spread skip avoids unnecessary object creation on every parent row. The method extraction reduces closure allocation overhead. Together, these reduce per-call overhead in the most frequently called join method.

Testing

• All 336 existing join tests pass (fetch/push/sibling + flipped variants)

[vercel]

Someone is attempting to deploy a commit to the Rocicorp Team on Vercel.

A member of the Team first needs to authorize it.