[hardening] DocxParser: preserve line breaks, tabs, and soft hyphens (stop using InnerText)

[MTCMarkFranco]

Enhancement

DocxParser.ProcessParagraph calls para.InnerText, which concatenates all descendant text without preserving:

• Explicit line breaks (<w:br/>) — collapsed.
• Tabs (<w:tab/>) — emitted as empty string.
• Soft hyphens / non-breaking hyphens (<w:softHyphen/>, <w:noBreakHyphen/>).

Impact:

• Multi-line addresses, signature blocks, and bulleted clauses-with-line-breaks become a single run-on string, breaking sentence-boundary heuristics in selectors.
• Tab-separated content (e.g. tabular paragraphs that aren't tables) loses field separation.

Acceptance criteria

• Replace para.InnerText with a run-walker that emits \n for Break, \t for TabChar, - for SoftHyphen/NoBreakHyphen.
• NormalizeInlineParagraph updated (or a new normalizer) to keep intentional \n while still collapsing whitespace runs inside a single visual line.
• Unit tests covering each token type.
• Existing parser tests still pass.
• Engine genericity guard still passes.

Out of scope

• Hyperlink target preservation (separate concern).