Add tests for a fragile early-stage pipeline

[karandaryanani]

Context

This is a new, somewhat fragile implementation. The classifier pipeline, evaluator, and splitter all work, but they could use proper testing to catch regressions as things change, especially with the backend and model swaps being planned (#2, #5).

Scope

• Unit tests for core functions (build_query, retrieval, reranker, splitter, evaluator)
• Integration test for the full classify_row() path
• Edge cases: short or garbage input, non-English text, and empty fields