Add ramp-based static INL/DNL analysis

[chenzc24]

Summary

• add ramp-code histogram helpers for static DNL/INL analysis
• expose analyze_inl_from_ramp and compute_inl_from_ramp through the public API
• add a debug example plus user-guide docs for ramp-vs-sine assumptions and endpoint baselines
• cover ideal ramps, missing codes, injected DNL, endpoint handling, validation, plotting, and guide snippets

INL/DNL convention

• DNL is reported per output-code width in result['code']
• INL is reported per transition level in result['transition_code'], so len(result['inl']) == len(result['dnl']) + 1
• raw transition INL satisfies result['raw_inl'] == [0, cumsum(result['dnl'])], matching the standard relation INL[k] = sum(DNL[:k])
• ramp INL defaults to endpoint='endpoints', so the endpoint-INL curve starts and ends at zero
• endpoint='fit' remains available for best-fit INL, and endpoint='none' remains available for raw transition INL

Tests

• uv run --with pytest pytest tests/unit/aout/test_inl_from_ramp.py -q
• uv run --with pytest pytest tests/integration/test_user_guide_skill_examples.py -q
• uv run --with pytest pytest tests/unit/aout -q
• uv run --with pytest pytest tests/unit/test_skill_cli.py -q
• uv run python src/adctoolbox/examples/04_debug_analog/exp_a33_inl_from_ramp.py
• git diff --check

Closes #49
Closes #50