fix: retry hosted compute updates after failed apply

[alongubkin]

Summary

• Resubmit the requested compute target for hosted deployments in pending or retryable lifecycle states, even when it equals the persisted desired settings.
• Keep the equality fast path for fully running deployments.
• Cover failed, pending, setup, provisioning, and updating lifecycle retries.

Why

Platform persists desired settings before the deployment engine applies them. After a failed apply, the desired target may already match the caller while the cloud still has the previous capacity. Re-submitting the target lets Platform’s idempotent operation queue repair the incomplete update.

[greptile-apps]

Greptile Summary

The PR changes hosted compute updates so equal persisted targets are resubmitted while a deployment is in any supported non-running lifecycle state, allowing interrupted applies to be retried while preserving the equality fast path for running deployments.

• Adds a lifecycle-aware predicate for deciding whether compute settings must be submitted.
• Covers active, pending, and retryable hosted deployment states with a focused unit test.

Confidence Score: 5/5

The PR appears safe to merge with no concrete correctness or security issues identified.

The new predicate preserves changed-compute updates for running deployments and resubmits equal targets only for lifecycle states already admitted by the existing hosted-compute status guard.

Important Files Changed

Filename	Overview
crates/alien-deploy-cli/src/commands/up.rs	Broadens hosted compute submission only within the existing supported-status gate and adds tests for running, in-flight, and failed lifecycle states.

_{Reviews (1): Last reviewed commit: "fix: retry hosted compute updates after ..." | Re-trigger Greptile}