Replace lock with threadvar for v1/v6 state, bump to 0.2.1

Use {.threadvar.} instead of std/locks for per-thread timestamp,
clock sequence, and node ID. Eliminates lock contention, removes
std/locks dependency. Per-thread monotonicity is sufficient per
RFC 9562.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Fix thread safety, hide internal API, bump to 0.2.0

- Fix race condition: move initLock to module scope, protect lazy
  init with the lock to prevent double-init from concurrent threads
- Use gregorianEpochOffset constant instead of hardcoded literal
- Hide setVersionAndVariant from public API via selective exports
- Use dashPositions constant consistently in parse formatting

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>