[Loops] Loop Goal integration fixtures discard required task claim tokens #417

Summary

The shared daemon integration fixture for managed Goal sessions claims loop-bound task runs and then starts them without forwarding the returned claim token. Current task lifecycle validation correctly rejects those starts with task: invalid claim token, leaving five subtests in TestLoopGoalManagedRuntimeIntegration red on main.

This is a test-fixture ownership bug, not a request to relax runtime claim validation.

Reproduction

On current main:

mise exec -- go test -tags=integration ./internal/daemon \
  -run '^TestLoopGoalManagedRuntimeIntegration$'   -count=1 -v

Five managed-runtime cases fail while constructing the fixture, for example:

StartRun("taskrun-goal-managed-terminal") error =
store: transition task run starting transaction:
store: execute sqlite write callback:
task: invalid claim token

Root cause

seedDaemonTaskRunLifecycle calls ClaimNextRun for loop-bound workers, assigns only claim.Run to its local state, and discards claim.ClaimToken. It later calls manager.StartRun without that token.

Three other daemon E2E fixtures were updated to use their claim token, but the shared managed-Goal fixture still follows the stale path.

Expected behavior

The fixture must carry the exact token returned by the successful claim into the claimed-to-starting transition. Production validation must remain fail-closed for missing or incorrect tokens.

Acceptance criteria

• seedDaemonTaskRunLifecycle (or its canonical replacement) forwards the exact claim token for loop-bound runs.
• Direct daemon admission remains tokenless only where the existing contract permits it.
• A regression test proves the helper cannot start a claimed run after dropping or replacing the token.
• TestLoopGoalManagedRuntimeIntegration passes in full.
• The three previously repaired daemon E2E scenarios remain green.
• No production claim-token validation is weakened.