solution-lod run blocks with 'Could not schedule implement' — implement sessions die after 900000ms inactivity

Environment

• opencode 1.18.20, macOS (darwin)
• graph: solution-lod, model glm-5.3

What happened

Run d108ca21-b2f7-4db1-9b9b-034f8191b467 (graph solution-lod) ended with status failed, phase blocked, result:

The solution network is blocked: Could not schedule implement for r2.

Three consecutive implement activations for region r2 failed with:

a3: OpenCode session ses_fd490bf28ffeV8EZOkVciHyNaO was inactive for 900000ms
a4: OpenCode session ses_fd4797281ffeHp2GUayX2sTF8r was inactive for 900000ms
a5: OpenCode session ses_fd462058effelopr1Wgdxn9Jur was inactive for 900000ms

The solution network itself was healthy (r1 collapsed, r2 actionable with a selected candidate) — the run died purely because spawned implement sessions hit the 15-minute inactivity timeout. After 3 retries the scheduler gave up and permanently blocked the run.

This is a continuation of the same failure mode seen in earlier runs (e.g. run 5d9efb9e died at implement:r6 scheduling for the same reason).

Expected

• Implement sessions should stay alive while work is pending, or
• The graph should reschedule/retry beyond 3 attempts instead of permanently blocking the run.