Continue stopping remaining run processes after per-target failures #719

Summary

maco agents stop --all --run-id <id> aborts at the first target failure, leaving later selected processes unattempted and omitting earlier successful results from CLI output.

Evidence and reproduction

On main fa0c79b9ae38eca2c59b2b1911becfe1a6b3c9ab, src/agent_lifecycle.rs:938-943 delegates the selected run processes to stop_records. At lines 991-998, errors from either termination (993-994) or registry removal (995) return immediately from the loop. AgentStopReport at lines 674-677 only represents successful entries. src/cli.rs:3017-3034 propagates the error before printing the report.

Use three registered fixture processes and inject a termination or registry-update error for the first or middle entry. Later entries are never attempted, and the CLI emits no structured partial result. Existing single-target tests do not exercise this case.

Acceptance criteria

• Attempt each independently selected identity even when another target fails.
• Preserve PID/start-time checks; uncertain or reused identities must never be signalled.
• Report each target as confirmed stopped, already gone, or failed with a typed reason, distinguishing termination from registry-bookkeeping errors.
• Remove registry identities only after confirmed termination/absence; retain unresolved records for safe retry.
• Emit the complete JSON report and return nonzero when any target remains unresolved.
• Test first/middle failure, mixed success, safe retry, stale identity and bookkeeping failure.

Scope and verification

Source inspection only; no host process or provider was interrupted. Closed #8 introduced whole-run stop, and #123/#644 handle separate identity/disappearance races. No open Issue/PR covers batch error aggregation.