Sprint 6: add opt-in robust variance-ratio diagnostics and release v0.2.0

[LautaroParada]

Motivation

• Add an opt-in robust variance-ratio family (AVR, WBAVR, Chow-Denning-style max test) while preserving the existing v1 weak-form battery behavior and public API.
• Introduce minimal internal refactor to separate diagnostic helpers from orchestration for easier extensibility and to support efficient bootstrap math on pre-normalized arrays.
• Provide deterministic, well-tested implementations and documentation for v0.2.0 without adding heavy dependencies or rolling support for the new robust tests.

Description

• Added RobustVRConfig and extended BatteryConfig with robust_vr, implementing validation rules in __post_init__ and exporting RobustVRConfig from the package root (src/variance_test/models.py, src/variance_test/__init__.py).
• Refactored internal helpers into a new src/variance_test/diagnostics/ subpackage and moved classic helpers (vr_classic.py, ljung_box.py, runs.py, arch_lm.py) out of battery.py while keeping battery.py as the orchestrator.
• Implemented robust VR internals and pure math helpers in diagnostics/robust_vr.py, including _compute_vr_value_from_log_prices, _compute_heteroskedastic_z2_from_log_prices, automatic q selection (AVR), wild-bootstrap WBAVR using Rademacher weights, and a Chow-Denning-style max test with Sidak-normal calibration (chow_denning).
• Integrated robust tests into run_weak_form_battery as opt-in, appending avr, wbavr, and chow_denning in the required order and adding multiple_testing["robust_vr_summary"] only when robust VR is enabled; v1 outputs and rolling payload are unchanged when not enabled (src/variance_test/battery.py).

Testing

• Ran the full test suite with pytest -q, which completed successfully: 99 passed with a small number of non-critical warnings relating to numerical edge cases.
• Added comprehensive unit tests for the robust features in tests/test_robust_vr.py covering config validation, pure-math parity with EMH().vrt, deterministic bootstrap behavior, returns/log_prices equivalence, edge cases, and synthetic IID/AR(1) behavior.
• Verified public API expectations in the updated tests/test_public_api.py and ensured RobustVRConfig is exported and present in __all__.
• Attempted python -m build in this environment but packaging build could not be executed because the build module is not available and network-restricted installation of that tool failed; this did not affect test success.

---

Codex Task