v0.11.1

Duplicate data in stream responses

A stream could emit the same chunk twice when workers were briefly unavailable, corrupting the response body.

• Duplicate chunks fixed. When scheduling a chunk failed because every worker holding it was in a short backoff, the portal both queued the error and re-queued the chunk. Once the backoff expired the chunk ran twice, and both results were written to the response.
• Readiness failures are explained in the logs. The /ready probe now reports why it isn't ready — no workers known, too few connections, or shutdown — logged once on each transition rather than on every check. (#119)

Full Changelog: v0.11.0...v0.11.1