feat: support runnable restart cycles

[mrcrgl]

Summary

• make RuntimeGuard fanout resilient across ticker generations by retrying delivery when a ticker disappears between readiness check and send
• require fanout to wait for a live ticker sender (!is_closed()), not just presence of a sender slot
• add restart-focused regression tests:
  • test_runtime_guard_supports_restart_after_ticker_drop
  • test_runnable_can_restart_start_shutdown_start
• update changelog under Unreleased -> Fixed

Why

Issue #18 requires Runnable instances to support start, shutdown, start cycles. The runtime fanout could terminate when control was sent while a previous ticker sender was closed, which broke restart reliability.

Validation

• cargo fmt --all -- --check
• cargo clippy --all-features -- -D warnings
• cargo test --all
• cargo test --no-default-features

Closes #18