Use a balanced tree instead of unbalanced one to prevent recursion error in create_match_filter

[koenvo]

Use a balanced tree instead of an unbalanced one to prevent recursion error in create_match_filter

Rationale for this change

In the create_match_filter function, the previous implementation used functools.reduce(operator.or_, filters) to combine expressions. This approach constructed a right-heavy, unbalanced tree, which could lead to a RecursionError when dealing with a large number of expressions (e.g., over 1,000).

To address this, we've introduced the _build_balanced_tree function. This utility constructs a balanced binary tree of expressions, reducing the maximum depth to O(log n) and thereby preventing potential recursion errors. This makes expression construction more stable and scalable, especially when working with large datasets.

# Past behavior

Or(*[A, B, C, D]) = Or(A, Or(B, Or(C, D))

# New behavior
Or(*[A, B, C, D]) = Or(Or(A, B), Or(C, D))

Are these changes tested?

Yes, existing tests cover the functionality of Or. Additional testing was done with large expression sets (e.g., 10,000 items) to ensure that balanced tree construction avoids recursion errors.

Are there any user-facing changes?

No, there are no user-facing changes. This is an internal implementation improvement that does not affect the public API.

Closes #1759
Closes #1785

[Fokko]

I think we also want to apply this in the And situation :)

[Fokko]

I think this is pretty elegant, thoughts @kevinjqliu?

I expressed some concerns with in #1783 around performance. I have played around with the singledispatch stuff in the past, and it was always crazy fast because the type dispatching is pushed down into the C-layer of Python.

[Fokko]

Moving this forward since I think it is a great idea in general :)

[Fokko]

Thanks @koenvo