Port 'Adjust placement of paragraph markers' from machine.py

[Copilot]

Ports machine.py#298 — after alignment-based placement of paragraph markers, apply small boundary adjustments to produce more natural splits (e.g. keeping a trailing comma with its sentence rather than letting it open the next paragraph).

New: SegmentBoundaryAdjuster

Two new classes in SegmentBoundaryAdjuster.cs:

• TokenRejoiner — reconstructs token lists into strings with correct punctuation spacing (no space before ,/./closing quotes, no space after opening brackets/quotes).
• SegmentBoundaryAdjuster — adjusts a segment boundary by:
  • Moving prohibited segment-starting characters (, ; . ? ! closing quotes/brackets) from the head of the next segment to the tail of the current one
  • Moving prohibited segment-ending characters (opening brackets/quotes) from the tail of the current segment to the head of the next one
  • Correcting late sentence starts (capitalized words that crossed the boundary too early)
  • Correcting early sentence ends (words + terminal punctuation that crossed the boundary too late)
  • AdjustTokenizedSegmentPairBoundaries(int boundary, IReadOnlyList<string> tokens) — token-index variant used by the handler

Change: PlaceMarkersUsfmUpdateBlockHandler

After PredictMarkerLocation, paragraph markers now go through AdjustTokenizedSegmentPairBoundaries before their string index is resolved:

// If inserting a paragraph marker, make small adjustments to place it in a more natural location
if (element.Type == UsfmUpdateBlockElementType.Paragraph)
{
    adjacentTargetToken = _segmentBoundaryAdjuster.AdjustTokenizedSegmentPairBoundaries(
        adjacentTargetToken,
        targetTokens
    );
}

Before: alignment places \p before , → paragraph opens with , y esta prueba…
After: comma stays in the preceding paragraph → Este texto está en inglés, / \p y esta prueba…

---

This change is