Report-back reaches the parent exactly once across a server restart #94

What: A child thread's report-back reaches its parent exactly once even across a Chromeria server restart.
Why: Report-back state is held in memory and restored from the parent's newest activities, so after a restart a report can be lost or repeated (childReportState.ts documents the gap).
So what: A worker implements it; the planner reviews and merges.

Parent

#92

Outcome

After a server restart, every child's next finished turn is reported to its parent once, and earlier reports are not resent.

Elon record

• Requirements and who asked: Prism routes delegated work through Chromeria's thread tools; retire the Python runner #92 acceptance: report-back must survive a restart so the dispatcher and planner are always woken.
• Deleted: the runner's job database as the durability layer.
• Bottleneck: dispatchers wait on report-back; a lost report stalls the job silently.
• Checked myself: apps/server/src/mcp/toolkits/threads/childReportState.ts header describing the activity-window gap.

Acceptance criteria

• A child with reportBack on or off keeps that setting after a restart, regardless of how many parent activities followed.
• A child's reply already reported before the restart is not reported again; a reply finished during the restart is reported once.

Non-goals

• Changing what report-back sends.

Blocked by

None

Required proof

• Tests at the threads toolkit seam that simulate restart beyond the activity window; CI green on the exact head; independent review.