feat: restore agents on demand

[mattjohnsonpint]

• Restore suspended agents on demand as needed rather than on startup
• Simplify agent suspension process during clean shutdown (and parallelize)
• Remove extra wait after spawning agent (not needed, since next message will be on same cluster node)

This should dramatically improve the thrashing and memory consumption we've been seeing whenever we roll out a new version to the cluster.

[Copilot]

Pull Request Overview

This PR enables on-demand restoration of suspended agents, simplifies suspension during shutdown with parallel execution, and removes unnecessary cluster sync waits.

• Add debug logging for topic subscriptions
• Implement retry logic in SendAgentMessage to restart missing agents on demand
• Remove automatic agent restoration on startup and inline cluster sync waits; parallelize agent suspension during shutdown

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
runtime/actors/subscriber.go	Log debug message when subscribing to a topic
runtime/actors/agents.go	Remove startup cluster sync wait; add retry loop with on-demand actor spawning in SendAgentMessage
runtime/actors/actorsystem.go	Strip out restore-on-startup logic and helper; inline cluster sync wait; parallelize suspension
CHANGELOG.md	Add entry for v0.18.6 with “restore agents on demand” feature

Comments suppressed due to low confidence (2)

runtime/actors/agents.go:205

• The new retry logic in SendAgentMessage isn't covered by existing tests. Adding unit tests for scenarios like actor restart and eventual success would help ensure this behavior remains reliable.

	const maxRetries = 3

runtime/actors/actorsystem.go:176

• The shutdown process no longer stops subscription actors after suspending agent actors, which may cause subscription actors to remain running. Consider reintroducing the subscription actor shutdown loop after wg.Wait().

	wg.Wait()