refactor of testing suite

[gkennos]

Summary

Big time refactor of test suite

• A marker-driven CI gate - fails when required PostgreSQL or document-format tests are skipped or deselected
• Stronger seeded PostgreSQL coverage for graph and embedding results, runtime permissions, indexes, and timeout recovery
• Real assisted-planning tests, including sync/async behavior, malformed responses, provider failures, and MCP/REST paths.
• Real XLSX, DOCX, and PDF parsing tests, plus hermetic plugin discovery and plugin-present TUI coverage
• CI evidence for coverage and skips, plus base and extras wheel smoke tests

Checklist

• Applied exactly one label (breaking, feature, fix, dependencies, or chore)
• Tests pass locally (uv run pytest -q)
• Lint passes (uv run ruff check .)