chore: Improve autopilot liveness check

[fleupold]

Background

We currently consider the autopilot to be alive as long as it's able to produce sufficiently recent auctions in the solvable orders cache:

services/crates/autopilot/src/run.rs

Lines 70 to 76 in 79c7aac

	#[async_trait::async_trait]
	impl LivenessChecking for Liveness {
	async fn is_alive(&self) -> bool {
	let age = self.solvable_orders_cache.last_update_time().elapsed();
	age <= self.max_auction_age
	}
	}

However, this approach doesn't work in shadow mode (where we don't populate a solvable orders cache) and may miss the autopilot getting stuck inside a single run loop. This is since the cache is updated on a separate thread, which may still be doing fine while the system is still not propagating any auctions to solvers.

Details

The Runloop should populate a thread-safe "last auction processed" timestamp that can be shared with the Liveness struct. liveness should then be considered healthy if the timestamp is within a reasonable range from now (e.g. 3 minutes).

Note, that this behaviour subsumes the current check as we only kick-off a new auction if the solvable orders cache has update (cf. code).

Acceptance criteria

• Liveness checks are based on the last timestamp an auction runloop has successfully completed.
• The same liveness implementation is used across shadow and regular autopilot mode