Report all incompatible role overrides in one selection preflight #706

An explicit plan can pass maco supervise plan normalization and then reveal incompatible role overrides one role at a time during supervise run. On source 254cbcede869aa58befb162e6f1d2d4760b72400, a read-only source-inspection plan requested gpt-5.6-sol/high for Researcher and Auditor, with three lenses matching Auditor. The first run rejected Auditor. After changing Auditor and its lenses to xhigh, the next run rejected Researcher. Both stopped before any assignment or source command.

The saved selection inputs admitted gpt-5.6-sol/xhigh for both roles. These effort candidates came from task/authority priors and availability gating; they were not a statement that the provider cannot run high generally. Rejecting an explicit override without automatic fallback is expected.

Please collect and report every incompatible explicit role triple in one selection preflight, including the task/authority profile, rejection reason, and eligible candidate efforts from the same frozen inputs. Expose this through a selection-only validation path where inputs are available; unavailable live data must remain explicitly unknown. No workload launch, reservation consumption, fallback, or weaker authority should be needed to obtain the diagnostics.

Acceptance:

• A plan with incompatible Researcher and Auditor overrides reports both in one validation result.
• Distinguish model availability, effort compatibility, and authority/risk/class rejection.
• Validate custom lens triples against the selected Auditor in the same result.
• Preserve explicit-override refusal, frozen input provenance, and all dispatch/authority gates.
• Compatible plans preserve current selection behavior.

Relevant code: src/supervise/selection_bridge.rs (debug_override_for_role, runtime_catalog_from_priors, role task-profile construction, custom lens validation) and src/selection/selector.rs (override rejection). Related: #149, #665. This follow-up concerns diagnostics, not new judgment authority or a model default.