fix: qualify current Pandas producer workflows for 0.2.0 - #171

Release-readiness repair

Closes #170. Correct the independently reproduced installed-qualification gap and prepare 0.2.0 without publication.

• Handle only the complete Series.dt.weekday deprecation message from PMC NYSE schedule construction. The project-wide default remains warnings-as-errors. A real direct-Pandas caller must still raise, and the regression fails under broad message suppression.
• Replace the arbitrary warning-filter count cap with an actual deprecated datetime call that must raise; broad suppression makes this control fail too. Existing explicit bans on global category suppression remain.
• Expand the nine-environment installed matrix to 852 input, callback/history, Pipeline, identity, lifecycle, control and accounting cases, requiring zero skips. The qualification cohort deliberately uses current Pandas. Include PyArrow 26 or later: its upstream GH-51302 fix avoids deprecated Pandas index conversion in the Arrow-present legacy handoff. Runtime dependency floors are unchanged.
• Assert the original captured UTC sample instants alongside unchanged financial assertions. Generate 0.2.0 notes from the structured changelog and update the migration guide.

Actual evidence

• Fresh supported Python 3.12 environment with the reviewed main wheel: all 13 reported PMC producer cases fail before the repair. The corrected policy passes those unchanged cases plus the caller-warning control.
• PyArrow 25.0.1 reproduces the additional timezone-index deprecation through the actual Diagnostic legacy workflow. PyArrow 26.0.0 passes without any new warning filter.
• Local stable 0.2.0 wheel on Python 3.12 and independently installed sdist on Python 3.14: each passes all 852 selected cases, zero skips, both without Arrow initially and with the corrected Arrow-present cohort.
• Complete supported source suite with Pandas 3.1/PyArrow 26: 3,271 passed, nine optional cross-engine skips and five benchmark deselections. All coverage gates and before/after dependency checks pass.
• Stable-version artifacts reproduce byte for byte; content/Twine checks, installed docs examples, strict docs identity, 11,890 content links/52 external destinations, Ruff/format/ty and normal hooks pass locally.

No runtime module changed. No benchmark or licensed authentication was rerun. Final-head hosted checks and the final merged-source stable candidate must pass before release readiness can be reconsidered. Ordinary source merge is separate from publication: no release, tag, documentation deployment or licensed-gate bypass is authorized here. The pending 0.1.13 candidate remains preserved.

AI assistance

OpenAI Codex assisted with reproduction, upstream source review, compatibility handling, regression controls, installed qualification and release preparation. The maintainer remains responsible for the implementation and reported evidence.