Bump version to 0.1.0a6

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Bump version to 0.1.0a5

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Major token efficiency and architecture improvements

- Rewrote /watchdog-ingest to use per-document Agent subagents with parallel
  batches of up to 5; orchestrator context stays flat regardless of batch size
- Replaced shingle-based near-dup storage with MinHash (128 hash functions);
  added watchdog near-dup --summary flag to extract decision without loading
  full minhash arrays into model context
- Added file locking to write_vault so concurrent subagent writes serialize
  safely without registry corruption
- Added CLI commands for ingest subagents: entity-index, queue-status,
  is-duplicate, validate-extraction
- Moved domain knowledge skills to per-vault .claude/commands/records/;
  added watchdog refresh-skills command replacing setup --skills-only
- Removed confidential material check from ingest pipeline
- watchdog status and watchdog obsidian now infer current project from cwd
- Prune OS junk files (.DS_Store, Thumbs.db, desktop.ini) before removing
  empty _INCOMING/ subdirectories
- Per-project Claude Code allowlist in .claude/settings.json scoped to exact
  commands the ingest pipeline needs
- Updated README, GETTING_STARTED, and INSTALL accordingly

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Bump version to 0.1.0a3

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Bump version to 0.1.0a2

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Fix skill count to 34 in README

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>