Release v1.10.0

This release delivers major robustness improvements, cross-platform parity, and a highly informative feedback
layer designed for LLM-driven document edits. It fully resolves the findings from GitHub Issue #23, hardens the
MCP Email client experience, and introduces safe preview execution capabilities.

Highlights

1. Redlining Engine Core Fixes (Issue #23 Compliance)

• BUG-23-1 (From-Scratch Comments Namespace): Resolved critical XML validation failures in Microsoft Word by
  ensuring word/comments.xml explicitly declares the xmlns:w14 (and w15/w16cid) namespaces when created from
  scratch.
• BUG-23-2 (Formatting / Italic Inheritance): Inserted runs (w:ins) now explicitly default to regular/roman
  formatting (empty run properties) instead of silently inheriting formatting like italics from surrounding anchor
  runs.
• BUG-23-3 (Punctuation Tokenizer Warning): Added a proactive tokenizer warning to alert callers when an edit
  anchor contains split-prone punctuation (e.g. _, -, or boundary transitions) that can cause mid-token
  alignment splits.
• BUG-23-4 (Multi-Paragraph Target Boundary Safety): To prevent silent structural corruption, both Python and
  Node engines now uniformly and safely reject any multi-paragraph target_text with identical, actionable error
  messages.
• BUG-23-5 (Tracked Deletions & Ambiguity Checks): Fixed unique matching by ignoring content inside
  tracked-deleted (w:del) elements during uniqueness and ambiguity validation checks.

2. Advanced Feedback & Preview Layer

• dry_run Mode: Introduced a no-write preview option (dry_run: true) to process_document_batch across
  both engines to simulate edit outcomes entirely in memory.
• Rich Semantic Reporting: Replaced bare execution tallies (Edits: N applied) with detailed per-edit
  feedback (apply status ✅ applied / ❌ failed, targets, warnings, and accepted-text previews).
• Preview Engine Correction: Solved a critical regression where the preview simulator's crude substring
  splicing caused prefix duplication and character eating. The simulator is now fully aligned with the actual
  writer's output on both engines.
• accept_all_changes Optimization: Patched memory/data leaks and finalized behavior during clean-up and
  finalization operations.

3. MCP Email Parity, Hardening & UX

• Achieved full behavioral parity between Python and Node implementations of the mcp-email tool (quote
  stripping, attachment size caps, stale-ID cleanups, robust timeouts, HTML stripping, and normalized local path
  parameters).
• Added exhaustive unit/integration tests for email tools (email.test.ts / test_email_tools.py).

4. Testing & CI Parity Validation

• Added comprehensive cross-platform, multi-engine regression suites matching both implementations to
  byte-identical output standards.
• Hand-crafted stripped XML fixtures (at_23_1_fresh.docx) to guarantee strict "no pre-existing comments"
  coverage.

What's Changed

• Add issue-23 regression detection tests (Python + Node + cross-platform) by @mkorpela in
  #24
• Topic fix all by @mkorpela in #25
  Full Changelog: v1.9.0...v1.10.0