refactor(compute-providers): reuse EC2 runner clients

[edersonbrilhante]

Description

Depends on #5308.

Introduces an Ec2RunnerClient that owns a reusable EC2 client and exposes request-bound EC2 runner operations through forRequest({ signal }). The control-plane registry creates the client once for the warm Lambda execution environment, and the pool, scale-up, and scale-down capabilities share those operations while keeping cancellation scoped to each request.

The EC2 capability boundary now distinguishes resource operations from provisioning operations, uses consistent capability naming, and removes the legacy default-operation wrappers. Launch-template block-device lookup uses the same bound EC2 client, and runner creation is named separately from runner configuration.

Test Plan

• Ran Prettier, ESLint, and the control-plane TypeScript type-check.
• Ran the complete compute-provider suite: 286 tests passed.
• Ran the complete control-plane suite: 345 tests passed.
• Added three runner-client tests covering launch-template lookup with the bound client and request signal, rejection of an already-cancelled request before AWS is called, and isolation between cancelled and active requests sharing one client.
• Updated pool, scale-up, and scale-down tests to exercise the returned capability methods and their dependency wiring.
• Ran git diff --check.

Related Issues

None.

[github-actions]

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None