Ogma v0.5.0

Concurrency guard — messages handled in per-chat worker threads with one in-flight message per chat (a second is dropped with a notice), plus a global semaphore capping concurrent claude runs via OGMA_MAX_CONCURRENT (default 1, protects RAM on a Pi); a long run no longer blocks other chats. bin/setup --check — non-interactive doctor validating token/allow-list/model/CLI/service/skills. Startup token validation — one clear token OK / TOKEN CHECK FAILED log line instead of looping 401s. See CHANGELOG.md.