Fix budget exhaustion re-dispatch loop

[cursor]

Summary

• Preserve _turn_budget_exhausted across G1 stale-claim pruning so budget-exhausted active tickets are not re-dispatched on the next poll tick.
• Add a regression test covering the next-tick prune path after max_total_turns exhaustion with no budget_exhausted_state configured.

Type

• Bug fix
• Feature
• Docs
• Tests or tooling
• Refactor

Verification

/tmp/symphony-validate-venv/bin/python -m pytest tests/test_orchestrator_dispatch.py -q
# 87 passed in 1.55s

PATH="/tmp/symphony-test-bin:$PATH" /tmp/symphony-validate-venv/bin/python -m pytest -q
# 798 passed, 5 skipped, 2 warnings in 26.94s

PATH="/tmp/symphony-test-bin:$PATH" /tmp/symphony-validate-venv/bin/symphony doctor ./WORKFLOW.md
# FAIL: agent.kind=claude missing from PATH in this container
# FAIL: tracker.board_root /workspace/kanban does not exist in this checkout

Risk Notes

• Minimal in-memory guard change: stale _claimed entries still prune, but _turn_budget_exhausted remains sticky for the orchestrator process as intended by the budget guard.
• symphony doctor failures are environment/check-out prerequisites unrelated to this patch.

Checklist

• This PR targets dev.
• I added or updated tests for behavior changes.
• I updated docs/examples for user-facing changes.
• I did not commit secrets, logs, local run state, virtualenvs, or generated artifacts.
• CI is green or I explained any expected failure above.