Phase 4: cross-format Compare orchestration

[Paul Lizer (paullizer)]

Summary

• Add default-off native mixed-source Compare orchestration for one Source plus an initially single Target.
• Reuse the Phase 1 authorized manifest, capability partitions, bounded evidence envelopes, document-window analysis, existing tabular runner, and established pairwise/final comparison prompts.
• Preserve target order, source roles, native engine labels, source-version coverage, artifacts/citations, token aggregation, and partial target visibility.
• Fail closed with a clear limitation while the flag is off, preventing tables from being silently routed through narrative chunk analysis.
• Add Phase 4 functional coverage, documentation, release note, and bump the application version to 0.250.067.

Validation

• functional_tests/test_cross_format_compare_workflow.py
• functional_tests/test_tabular_document_actions_workflow.py (6/6)
• functional_tests/test_mixed_source_analyze_workflow.py
• functional_tests/test_mixed_source_manifest_contracts.py (13/13)
• Python compilation for all changed Python files
• Editor diagnostics clean
• Scoped broken-access-control and XSS scans passed
• CRLF-aware git diff --check passed

Rollout / Remaining Gate

• enable_cross_format_compare and enable_cross_format_compare_one_to_many both default to false.
• The initial enabled stage permits one Source plus one Target; one-to-many mixed targets remain staged off.
• Validated multi-file structured overlap/delta operations, full authorization and source-version runtime matrix, Foundry/streaming parity, and frontend Playwright coverage remain rollout-gate follow-up work.

Fixes#1059
Refs #1055
Refs #1056
Refs #1057
Refs #1058