v0.16.0

• Added incremental comparison against a fixed baseline and up to 16 earlier
  batches. CLI, MCP, HTTP, and dashboard show marginal gain and combined coverage
  without treating selected-test absences as regressions.

Breaking changes

• Replaced eight orchestration/review tools with read-only coverage_gaps and
  coverage_compare; removed legacy resources and source, run, command,
  artifact, topology, worktree, duplicate-test, and streaming dashboard routes.
• Added standalone report-file CLI queries; saved history remains optional.
• Rebuilt the dashboard around source locations and comparison outcomes.
• Added explicit source/build compatibility, selected-test absence semantics,
  optional producer receipts, and bounded function groups without source dumps.
• Legacy queued jobs are interrupted on open, never automatically restarted.
• Retained parsers, historical ordinary/composite data, and compaction.
• Upgrade the runtime and marketplace connector together, then start a new MCP
  session. Use the repository test runner directly; removed tools have no aliases.
• Incremental baseline reuse requires compatible measured source and
  instrumentation. It measures added coverage, not full-suite regression status.

Changed

• Consolidated onboarding into the README and migration into the user guide;
  architecture and fixture guidance now live in the contributor guide.
• Corrected restart/release instructions and restored symptom-based support.
• Added a tested Rust query example and shipped user guides with native archives.