Semantic search, AI summaries & security warnings

[MelbourneDeveloper]

TLDR;

Adds AI-powered semantic search and summaries to CommandTree using VS Code's Language Model API, with structured security warning detection via the Tool API.

Details

• Semantic search: Embeds commands using all-MiniLM-L6-v2, stores in SQLite (.commandtree/), ranks by cosine similarity with percentage scores
• AI summaries: Copilot generates human-readable summaries for discovered commands, displayed in tree tooltips
• Security warnings: Uses VS Code Language Model Tool API (toolMode: Required) for structured output — dedicated security_warning column in DB, ⚠️ icon on risky commands, warning text in tooltip
• Tagging rewrite: Tags now stored in SQLite junction table with exact command ID matching (no pattern matching). Managed through UI commands, not JSON config
• New discoverers: .NET projects (.csproj) and Markdown files (.md)
• Model selection: Automatic Copilot model selection with preference ordering
• Test improvements: Split semantic tests into summaries + semantic search files, removed fake tests (assert.ok(true)), added unit tests for similarity, embeddings, model selection
• Removed legacy: Deleted deprecated editTags command, pattern-matching tag config, tagconfig.provider.test.ts

How do the tests prove the change works

• E2E tests verify semantic search rankings, summary display in tooltips, security warning ⚠️ icons, tag filtering through the VS Code UI (black-box only)
• Unit tests isolate cosine similarity math, embedding serialization/deserialization, model selection logic, and DB operations
• All tests run in actual VS Code window via @vscode/test-electron against a temp fixture workspace