Validation: Batch 1B + Checkpoint 2.1

[imakris]

Cross-platform validation PR for the stacked Batch 1B and Batch 2 Checkpoint 2.1 candidate at GitHub head 3bac8f89d741097e735f678861fe943fe445f7ac. Delivery remains split across PR #17 and PR #18; do not merge this validation PR.

The candidate restores Qt's ordinary native fallback-surface path and default FreeBSD capacity/renderer settings, retains only the causal POSIX alignment normalization and regression, moves all normal phase-trace flushing outside measured frames while flushing every aggregate boundary, removes the dependency version constraint, and keeps the Windows extended-path adapter with a >300-character regression. It also contains the owner-approved architecture convergence register and subsequent review amendments; those documentation commits do not change runtime behavior.

Local exact-head gates are green: fresh Release configure/build, CTest 30/30, all direct Python helper suites, canonical Varinomics style (two existing advisory ternary warnings), pinned actionlint 1.7.12, and native D3D11 pixel/counter smoke. The retained local gate bundle is DIAGNOSTIC_PASS; calibration and owner approval have not started.

Fresh hosted validation is 8/8 green: Linux 29142137332, FreeBSD 29142137301, macOS 29142137307, and Windows 29142137324. All four text-enabled bundles independently recompute every declared artifact hash and retain clean equal source/build/runtime trees, clean dependency f5f0b120, native non-Null rendering (OpenGL/OpenGL/Metal/D3D11), nonuniform pixels, nonzero checksums, and complete phase traces.

A1 remains diagnostically red rather than closed: the workflows check out GitHub's synthetic PR merge commit in their manifests even though the run headSha and identical tree correspond to 3bac8f8, and the only full runner mode enters calibration without a pre-calibration terminal mode. A focused remediation unit is required before a fresh exact-source A1. This validation PR must not be merged.