Gracefully handle contested persistence lock when running multiple instances

[hugwi]

Problem

Running multiple OpenCode instances from the same directory fails with:

ERROR goal persistence is already owned by pid <PID> on <hostname>;
close the other OpenCode instance or configure a different stateFilePath

The acquirePersistenceLease call in ensureSessionLoaded (goal-plugin.js:3106) throws a hard error when another instance already holds the lock. This makes it impossible to run multiple instances from the same working directory — a common workflow.

Proposed Fix

Wrap the lease acquisition in a try/catch and return gracefully when the lock is contested. The session starts normally; goal persistence simply skips for that instance while the lock-holding instance continues to persist.

-      const lease = await acquirePersistenceLease(paths.stateFilePath)
+      let lease
+      try {
+        lease = await acquirePersistenceLease(paths.stateFilePath)
+      } catch (leaseError) {
+        if (String(leaseError?.message || leaseError).includes("goal persistence is already owned")) {
+          // Another instance holds the lock — continue without persistence.
+          return true
+        }
+        throw leaseError
+      }

File: src/goal-plugin.js, function ensureSessionLoaded, line ~3106.

Trade-offs

• The non-lock-holding instance runs without goal persistence (goals won't be saved to disk for that session).
• The lock-holding instance persists normally.
• This matches the existing pattern in acquireMigrationLease, which already retries gracefully on the same error (line ~1488).

Alternative

Allow configuring a per-instance stateFilePath via env var (OPENCODE_GOAL_STATE_PATH) so each instance gets its own state root. The plugin already supports this option, but it requires manual setup. The graceful-fail approach is a better default UX.