Finalize pytest audit for #36

[chenzc24]

Summary

Final stack item for #36. This keeps remaining broad diagnostic sweeps out of the default pytest run while preserving them under pytest -m slow, and strengthens the remaining plot smoke coverage with structural assertions.

Changes:

• mark broad window sweep diagnostics as slow and assert plot axes/lines/labels plus finite metrics
• mark the noise-shaping signal-generator dashboard as slow and assert subplot/result structure
• mark heavy NTF analyzer diagnostics as slow while keeping the lightweight plotting test in default pytest with NTF response assertions

This branch is stacked on #47 in commit order. Together with #45-#47, this addresses the issue owner's follow-up checklist: deterministic test RNGs, stronger plot assertions, and a slow-test audit for broad exploratory sweeps.

Verification

• uv run --with pytest pytest tests/unit/spectrum/test_window_sweep.py tests/unit/siggen/test_noise_shaping_siggen.py tests/unit/oversampling/test_ntf_analyzer.py -q
• uv run --with pytest pytest -m slow tests/unit/spectrum/test_window_sweep.py tests/unit/siggen/test_noise_shaping_siggen.py tests/unit/oversampling/test_ntf_analyzer.py -q
• uv run --with pytest pytest tests/unit/spectrum tests/unit/siggen tests/unit/oversampling -q --durations=20
• uv run --with pytest pytest --collect-only -q
• git diff --check

Part of #36.