Fix Rust parity gaps and add differential evidence

[nicosuave]

Fixes fanout, default-time, population ownership/policy, and window alias mismatches between the Python and Rust engines. Exact dotted graph identities now retain their ownership and visibility checks; native cumulative identifiers are quoted; Python evaluates source-local formulas after missing-source counts/defaults are restored. Unsupported Redshift approximate distinct windows fail explicitly.

Adds expected-result regressions, minimized replay cases, a seeded DuckDB differential runner, and an auditable inventory of shared-suite exclusions and helper coverage. Raises the DuckDB minimum to 1.4.4 for cumulative NULL-date correctness and makes date-bucket expectations compatible with DuckDB 1.5. Fixes Snowflake emulator dependency resolution and date-only browser filter bounds in non-UTC timezones.

Python remains the default. The inventory still contains 54 real-gap records, including unproven native coverage; these are not 54 confirmed product defects. A default switch remains separate work.

Validation on Python 3.13 / DuckDB 1.5.6 with a rebuilt native engine:

• Both full shared lanes: 11,724 passed, 147 skipped, 89 deselected, and two funnel date-expectation failures. Those assertions were subsequently repaired; the complete funnel module passed 45 tests. Python coverage was 79%, above the 76% gate. The full lanes were not repeated after the final small fixes.
• Final calculation follow-ups: 2,913 semantic-conformance passes with only those same two date assertions failing before their repair; 958 metric/query tests passed; all 520 generated model/query inputs agreed across engines, and all six replay fixtures passed.
• 657 Rust unit tests, 56 population/visibility regressions, 12 browser tests, and Snowflake emulator tests for both engines passed. Date expectations were also checked against DuckDB 1.4.4.
• Repository lint/format, public API type checks, and diff checks passed. The broader platform, runtime-feature, wheel, and external-service matrix remains CI's validation scope.

Follow-up verification: all main CI and integration checks passed on 1d75fa6f; the only remaining failure was live Yardstick fixture timestamp formatting. 35acb68d repairs that adapter and adds regressions without relaxing time-value checks. The complete live and vendored replay module passed 27 tests against Yardstick 0b5fc827321b71b7bb6379d9259fe8b2ea243a79. Final-head CI is pending.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-07T00:08:04.650186Z	35acb68	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.