Trim defensive internal invariant checks

[adrunkhuman]

Summary

• remove low-value defensive checks from internal scoring and city-ranking runtime objects that are constructed only by our own code
• drop tests that intentionally corrupt those internal objects only to assert a specific validation exception
• remove stub-repository fixture tests that only proved returned objects had expected Python classes
• keep real validation at trust boundaries such as HTTP input, DuckDB reads, build-time database validation, and environment parsing

Why

This started from #91, but the useful cleanup is not an exception-class swap. The problem was that some tests and constructors treated programmer-controlled runtime objects as if they were hostile input. That added noise, froze implementation details, and made internal corruption paths look like supported behavior.

The cache/vectorization work in #39, #65, and #87 made these internal models more alignment-sensitive, so defensive guards were understandable while those paths stabilized. At this point the construction sites are narrow enough that normal NumPy/Python failures are sufficient if our own code corrupts an object. Boundary validation remains intact where malformed external data can actually enter.

I also did a broader pass for the same pattern. The remaining obvious brittle area is tests/test_app_shell.py, which over-specifies rendered markup and static script details. That is real, but it is a separate frontend contract-test cleanup rather than part of this internal runtime object audit.

Testing

• uv run pytest tests/test_scoring.py tests/test_climate_repository.py tests/test_score_service.py tests/test_heatmap.py -q
• uv run ruff check backend/scoring.py backend/cities.py backend/heatmap.py backend/climate_repository.py tests/test_scoring.py tests/test_climate_repository.py tests/test_score_service.py tests/test_heatmap.py
• uv run ty check backend tests/test_scoring.py tests/test_climate_repository.py tests/test_score_service.py tests/test_heatmap.py

Closes #91