Simplify the committor and improve settlement reliability

Nobody has claimed this yet.

## Goal

Make the committor easier to understand, change, and operate without putting settlement reliability at risk. Improve it in independently releasable stages, keeping pending work safe as the implementation evolves.

## Why

The committor has to get work onto the base layer and know what happened when something goes wrong. When ownership, retries, and recovery are spread across too many layers, it's harder to reason about those guarantees or diagnose a stuck intent.

The original [proposal in dev-kanban#203](https://github.com/magicblock-labs/dev-kanban/issues/203) called for a clean-sheet replacement. We're keeping the aim of a simpler service, but taking a more conservative route: improve the existing implementation step by step rather than concentrate the migration risk in one cutover.

## Scope

This umbrella covers the work needed to:

- Clarify service boundaries and simplify ownership and control flow.
- Move durable intent handling onto the V2 Engine foundation, so pending work survives recovery without duplication or loss.
- Make overload, retries, restarts, and dependency failures predictable.
- Improve health signals, operational controls, and diagnostics so problems are easier to spot and investigate.

Preserve ordering, idempotency, recovery, and confirmation guarantees as each part changes. Each stage needs a clear compatibility, validation, rollout, and rollback path; retire old paths only once their replacements are proven in production. Child issues will define the concrete changes and their release criteria.

The planned sequence is to begin after Validator V3 network-integrity work and proceed alongside the Chainlink rewrite. When capacity is constrained, this work takes priority over SDK V1.