Report which documentation files were excluded or lost during ingestion

[TheRealBecks]

A successful documentation build does not currently explain which source documents reached the package, which were intentionally excluded, or which were lost to read/parse failures. That makes missing documentation difficult to distinguish from an unsuccessful search.

There is already a useful skippedFiles count; this proposal extends it into actionable diagnostics rather than replacing it.

At upstream commit 149f112:

• findMarkdownFiles() applies filename, directory, language and ignore filters. Directory read failures are caught without a diagnostic.
• readLocalDocsFiles() deduplicates content and catches file-read failures without returning their paths or reasons.
• buildPackage() counts split/parse failures, but does not expose the affected paths or failure reasons. A file producing no sections is also different from a parser throwing.
• fix(context): only skip repo-meta filenames at the scan root #125 demonstrates the practical consequence: valid documentation named security.md can disappear while the build succeeds. That PR fixes a filtering rule; this issue makes similar omissions diagnosable throughout the pipeline.

Proposed behavior:

Return structured ingestion diagnostics alongside the existing build result. Distinguish intentional exclusions, duplicate content, unreadable files/directories, split/parse failures, files producing no sections, and successfully indexed documents.

The CLI should print a concise summary and offer detailed or machine-readable output with relative paths and reasons. Keep diagnostics in the result rather than requiring library code to log directly. Avoid flooding the report with unrelated source-code files when scanning a repository.

Acceptance criteria:

• A small fixture with exclusions, duplicate documents, a read failure, a parse failure, and an empty document produces distinct, inspectable outcomes.
• Git, ZIP, and HTML-index sources can report outcomes through the same vocabulary.
• The summary distinguishes discovered documentation, selected files, indexed documents, and indexed sections.
• Representative Markdown, HTML, and RST fixtures assert that expected headings and examples survive ingestion.
• Unexpected loss can fail an explicit strict validation mode; intentional exclusions remain distinguishable from failures.

Related: #125 and #137. The objective is extraction visibility and fidelity; retrieval ranking can be evaluated separately.

[TheRealBecks]

I will wait for your comment and #125 and #137.