ci: download en_core_web_sm model in PR workflow

[jexp]

Adds a step to download the en_core_web_sm SpaCy model (~12 MB) during the PR unit test workflow so that the SpaCy integration tests run properly instead of being skipped.

The step is conditioned on python-version != '3.14' to match the pyproject.toml guard (spacy==3.8.7; python_version < '3.14').

- name: Download SpaCy model
  if: matrix.python-version != '3.14'
  run: uv run --frozen python -m spacy download en_core_web_sm

Prerequisite for:

• feat: SpacyEntityRelationExtractor (LLM-free KG construction) #547 (SpacyEntityRelationExtractor — 3 integration tests)
• feat: HierarchicalTextSplitter — header-aware chunking with verbless sentence filtering #548 (HierarchicalTextSplitter — 2 integration tests)