[codex] move live context diffing into world state

[pakrym-oai]

Why

Model-visible settings diffing was split between the typed environment world state and ad hoc builders in context_manager/updates.rs. That left the in-memory world baseline incomplete, duplicated initial-context rendering, and meant environment changes during a multi-iteration turn could reach the next sampling request with stale context.

This PR is stacked on #29249.

What changed

• Add separate typed WorldState sections for model, permissions, collaboration mode, multi-agent mode, realtime state, and personality; each section owns only its topic comparison and fragment rendering.
• Pass each section an optional previous value, distinguishing an absent section from a persisted optional value without sentinel defaults or nested options.
• Return ordered ContextualUserFragments from world-state rendering and let the context consumer merge adjacent same-role fragments into messages.
• Build initial world context through WorldState::render_full(), including render-only environment subagent data and personality gating; retain model-switch and realtime-shutdown as transition-only overlays.
• Store the latest in-memory WorldState snapshot on ContextManager next to the durable TurnContextItem reference.
• Rebuild world state from the live environment snapshot before each sampling iteration and inject only its rendered diff.
• Seed or refresh the world baseline during resume reconstruction, compaction, and context-window replacement.
• Preserve the legacy single-environment rendering while allowing initial fragment ordering and message grouping to follow world-state assembly.

Testing

• just test -p codex-core build_initial_context world_state build_settings_update_items turn_iteration_injects_environment_diff_after_thread_environment_changes fork_startup_context_then_first_turn_diff_snapshot snapshot_model_visible_layout_turn_overrides snapshot_model_visible_layout_resume_with_personality_change multi_agent_mode_is_sticky_and_emits_only_on_change resume_compares_against_previous_effective_multi_agent_mode

[github-actions]

Closing this pull request because it has had no updates for more than 14 days. If you plan to continue working on it, feel free to reopen or open a new PR.