Refactor anonymiser UI to handle latest BSA version

[boscorat]

Summary

This PR refactors the anonymisation functionality to replace raw TOML text editors with intuitive form-based table inputs, addressing the breaking changes in the latest bank_statement_anonymiser package.

Changes

UI Improvements

• Never Anonymise tab: Single-column table for excluded phrases with Add Row / Remove Selected buttons
• Always Anonymise tab: Two-column table (Original Text | Replacement) with Add Row / Remove Selected buttons
• Auto-save behavior: Configs now save automatically on dialog close or "Run Anonymisation" button press (with 3-retry logic)
• Removed manual "Save Config" button
• Always show one empty row at bottom of tables for convenience

Code Refactoring

• Added NeverAnonymiseConfig data model for excluded phrases
• Added AlwaysAnonymiseConfig data model for forced replacements
• Both models support full TOML round-trip serialization
• Implemented retry logic: 3 attempts with 1-second delays between failures
• Removed ~40 lines of legacy migration code
• Removed dirty-state tracking and related signal wiring
• Simplified presenter logic significantly
• Improved docstrings for _run_anonymisation() and _on_dialog_finished()

Documentation Updates

• Rewrote "Configuring Anonymisation" section with table-based explanations
• Removed obsolete "Saving the config" section
• Updated "Config File Locations" to clarify two separate TOML files
• Improved feature overview to reflect new two-table model
• Added examples showing how each tab works

Testing

✅ All 129 tests pass
✅ Type check: 0 errors (pyrefly)
✅ Lint: All checks pass (ruff)
✅ Format: All files correctly formatted
✅ TOML round-trip: Verified with BSA system config files (155 phrases load correctly)

Files Changed

• docs/screens/anonymise.md — Documentation rewrite
• src/openstan/presenters/anonymise_presenter.py — Data models + presenter logic
• src/openstan/views/anonymise_dialog.py — Table-based UI

Architecture

• MVP separation maintained: View (tables) ↔ Presenter (auto-save) ↔ Data models (TOML)
• All widgets use Stan-prefixed subclasses
• Background worker thread for anonymisation (unchanged)
• Non-blocking UI with proper error handling

Resolves #79