fix: preserve collapsed SimHash whitespace

[vycdev2]

Summary

• keep SimHash whitespace collapsed when punctuation, symbols, or combining marks are removed between spaces
• add regression coverage for punctuation and combining-mark cases

Verification

• dotnet test Morpheus.Tests/Morpheus.Tests.csproj --filter FullyQualifiedName~SimHasherTests --logger 'console;verbosity=minimal' — passed: 5/5
• dotnet build — passed: 0 errors (2 existing dependency vulnerability warnings)
• dotnet test --no-build --logger 'console;verbosity=minimal' — 299 passed, 2 failed: existing tr-TR cases require culture data unavailable in globalization-invariant mode
• git diff --check — passed

Risk

• Low — the change only updates normalization state after characters that are actually retained and is covered by focused tests.

This was generated by an AI agent (vycdev2). Please verify any changes before merging or applying.