HtmlDocumentReader converted h1-h6 to Markdown but never set
structuralHints["has_headers"], so BasicDocumentParser fell back to
paragraph-based sectioning and large HTML was indexed as a single chunk.
Detect headings in AnalyzeSemanticStructure and set has_headers +
header_count, matching the PDF/Word/PowerPoint readers. +2 tests.

Also add the missing FileFlux.Core project to FileFlux.slnx so that
`dotnet pack FileFlux.slnx` produces FileFlux.Core.nupkg (pack closure
parity with CI's per-project pack).

Resolves AIMS upstream issue: HTML files indexed as one chunk. 0.10.11 -> 0.10.12