fix(controller): fill configured local runner capacity

[Bluu (Bluuok)]

With local_runner.maximum_size=60 or 128, the local controller still requests only 50 queuing/running rollouts. If those first 50 stay running, the remaining configured worker slots cannot be used and later queued work remains invisible.

Size the query limit to at least the configured pool capacity, preserving the existing minimum/default of 50.

Validation on Ubuntu 24.04 / Python 3.12 with locked CPU dependencies:

• python -m pytest -q tests/controller: 28 passed.
• Parameterized two-cycle regressions cover capacities 25, 50, 60, and 128. Unchanged main fails the 60/128 cases with only 50 active records; the patch passes all four. API responses honor the requested limit and process spawning is mocked.
• Ruff/format, Pyright, copyright headers, pre-commit, and diff whitespace passed.

No GPU training was run. This change is independent of terminal-process history cleanup in #618.

AI assistance was used to investigate, implement, and validate the fix. The final diff and regression behavior were reviewed before submission. The CLA check has passed; fork CI is awaiting maintainer approval.

[Copilot]

Copilot review overview

🟢 Approval recommended

The focused implementation correctly addresses capacity underutilization and includes representative regression coverage.

Review effort: Balanced
Findings: None

What changed in this PR

Updates local reconciliation to query enough active rollouts to utilize the configured worker pool.

Changes:

• Sets the query limit to at least the configured pool size while preserving the minimum of 50.
• Adds regression coverage for pool sizes 25, 50, 60, and 128.

File	Description
agentlightning/​controller/​local_reconciler.py	Scales the rollout query limit with local capacity.
tests/​controller/​test_local_reconciler.py	Verifies query limits and pool utilization.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.