fix: encode GAMS EPS as np.finfo(float).tiny (closes #39)

[elainethale]

Summary

• BREAKING: gdxpds.special.NUMPY_SPECIAL_VALUES[-1] is now np.finfo(float).tiny (~2.22e-308), not np.finfo(float).eps (machine epsilon, ~2.22e-16). EPS detection is exact equality on the sentinel, so a legitimate small float (e.g. 1e-200, machine epsilon) now round-trips as itself instead of silently mapping to GAMS EPS on write. Closes GAMS eps isn't the same as Numpy EPS #39.
• Migration: if you wrote numpy.finfo(float).eps to mean GAMS EPS, switch to numpy.finfo(float).tiny. gdxpds.special.NUMPY_SPECIAL_VALUES[-1] is always the canonical sentinel.
• Bumps to v4.0.0. Implements Item D of dev/modernization-wrapup.md, the final release in the modernization arc.

What changed

• src/gdxpds/special.py: NUMPY_SPECIAL_VALUES[-1] = np.finfo(float).tiny; is_np_eps and convert_np_to_gdx_svs use exact equality on the sentinel (no more |val - eps| < eps band).
• src/gdxpds/_gdxcc_engine.py _coerce_value_col: write-side eps_mask = arr == eps (exact equality); docstring updated.
• src/gdxpds/_transfer_engine.py _substitute_value_col: same exact-equality switch; docstrings now make explicit that gt.SpecialValues.EPS is the negative-zero bit pattern (distinguishable from +0.0 only via the sign bit, not ==).
• Tests: tests/test_specials.py, tests/test_engine_parity.py, tests/test_write.py all swap eps → tiny. New regressions: test_small_float_is_not_eps and test_write_parity_small_floats_not_eps (1e-200 survives the full 2x2 write x read engine matrix).
• Docs: doc/source/overview.md special-values table, "drop EPS" snippet, intro paragraph, and a new "v4.0.0 breaking changes" subsection in the migration section.
• CHANGES.txt: v4.0.0 entry.
• src/gdxpds/__init__.py: __version__ = "4.0.0".

Test plan

• Lint (ruff check + ruff format --check) clean.
• pytest tests passes on .venv-gams-34 (pandas 3.0).
• pytest tests passes on .venv-gams-49 (pandas 2.2, gamsapi 49.6.0).
• pytest tests passes on .venv-gams-51 (pandas 2.2, gamsapi 51.3.0).
• .venv-no-gams wheel build + gdxpds info succeed as expected.
• 1e-200 write/read round-trip preserved on both engines (covered by the new regressions).

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Fixes issue #39 by changing the numpy encoding of GAMS EPS from np.finfo(float).eps (~2.22e-16) to np.finfo(float).tiny (~2.22e-308), and switching EPS detection from a tolerance band (|val - eps| < eps) to exact equality on the sentinel. This is a breaking change bumping the package to v4.0.0; legitimate small floats (e.g. 1e-200, machine epsilon) now round-trip as themselves instead of being silently coerced to GAMS EPS on write. Changes are made consistently across the canonical mapping module, both I/O engines, tests (including new cross-engine parity regressions), docs, and changelog.

Changes:

• Replace np.finfo(float).eps with np.finfo(float).tiny as the canonical EPS sentinel in special.py, and switch all EPS detection (is_np_eps, convert_np_to_gdx_svs, gdxcc _coerce_value_col, transfer _substitute_value_col) to exact equality.
• Add regression tests (test_small_float_is_not_eps, test_write_parity_small_floats_not_eps) ensuring 1e-200 and machine epsilon survive the full write/read engine matrix.
• Update version to 4.0.0, plus CHANGES.txt and doc/source/overview.md (special-values table, EPS-drop snippet, new v4.0.0 migration subsection).

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/gdxpds/special.py	Canonical sentinel switched to tiny; is_np_eps and convert_np_to_gdx_svs use exact equality; docstrings updated.
src/gdxpds/_gdxcc_engine.py	Write-side EPS detection switched to exact equality; read-side docstring updated to tiny.
src/gdxpds/_transfer_engine.py	Same exact-equality switch for _substitute_value_col; docstrings clarify gt EPS is -0.0 bit pattern.
src/gdxpds/init.py	Version bump to 4.0.0.
tests/test_specials.py	Swap eps→tiny; add test_small_float_is_not_eps regression.
tests/test_engine_parity.py	Swap eps→tiny; add cross-engine small-float regression.
tests/test_write.py	Swap eps→tiny in test_write_known_value_columns.
doc/source/overview.md	Update special-values table, snippet, intro; add v4.0.0 migration subsection.
CHANGES.txt	New v4.0.0 entry describing the breaking change.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.