Sync agent 1.5.1 from monorepo

[DorwardTech]

1.5.1 — the agent stops going blind between polls while it is on the print server

The rule the whole results path is built around is that the print server is never contacted during a game, and every operation re-checks that immediately before it connects. But the check reads a value that only the telemetry poll ever writes, so it can never be more current than the last poll — and an idle venue polls every fifteen seconds.

A game starting just after an idle reading stayed invisible for up to that long, and a results drain or a whole-venue backfill would keep opening new fetches straight through the gap. An earlier release closed the version of this that could last for an entire multi-game drain; what was left behind it was the gap between two polls.

The agent now polls at the in-game rate for as long as it has print-server work in progress. Blind spot: fifteen seconds → five on the default settings. It cannot reach zero — an operation already in flight still runs to its timeout, which is a separate and known limitation — but it bounds how long after play begins a new one can start.

The cost is that telemetry is pushed at the in-game rate for that window too. A results drain is seconds; a backfill is minutes, and an operator asks for those one at a time.

Verification

• gofmt -l ., go vet ./..., go test -race ./..., go build ./... — all clean, run in this repo before pushing
• The sync was diffed against this repo first: nothing exists only here, so nothing was deleted, and the three hand-maintained overlay files (Dockerfile, .github/workflows/docker-image.yml, docker-compose.yml) differ only by their AGENT_VERSION stamp

After merge

Tag v1.5.1 to publish the image and the Windows archives.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Vxcyn1yt77BwnishNe2dvr

---

Generated by Claude Code