Add 8 Claude companion features

[drompincen]

Summary

• 8 new features to make JavaDucker a better companion for Claude Code on large codebases
• 6 new MCP tools: javaducker_summarize, javaducker_map, javaducker_stale, javaducker_dependencies, javaducker_dependents, javaducker_watch
• 2 enhanced tools: search results include line_start/line_end, uploads replace stale artifacts via incremental re-indexing
• HNSW index: pure Java O(log n) ANN search for semantic queries at scale
• 7 new Java classes, 13 modified files, 31 files total, 65/65 tests passing

Features

#	Feature	What it does
1	Line Numbers	Chunks track lineStart/lineEnd so Claude can Read(file, offset=N)
2	Incremental Re-indexing	Re-uploading a file replaces stale data instead of duplicating
3	File Summaries	Regex-based class/method/import extraction for Java, JS, TS, Python, Go, Rust
4	Project Map	Directory tree with file counts, largest files, recently indexed
5	Dependency Graph	Import parsing + resolution, dependency/dependent queries
6	Diff-Aware Search	Detects stale indexed files via mtime vs indexed_at comparison
7	Watch Mode	WatchService-based auto-indexing with debounce on file changes
8	HNSW Index	Pure Java approximate nearest neighbor search, wired into SemanticSearch

Test plan

• All 65 existing tests pass (including updated integration + unit tests)
• mvn compile clean, mvn test green
• Test constructors updated for new dependencies (IngestionWorker, UploadService)
• DuckDB PK constraint workaround verified (DELETE+INSERT for reindex)

🤖 Generated with Claude Code