fix: extend --job-array-* now tracks and reports per-instance reload failures

[scttfrdmn]

Summary

• extendJobArrayWithAudit discarded triggerReload's per-instance error with _ = triggerReload(&inst), unlike the single-instance runExtend path, which warns and prints a manual ssh ... sudo spored reload fallback on a reload failure.
• A caller extending a job array of N instances had no way to tell which instances actually picked up the new TTL on-instance vs. which are silently running on stale config until the next periodic tag refresh (5 minutes in production).
• Extracted the per-instance tag-write + reload loop into extendJobArrayInstances(instances, updateTags, reload, stderr) so the accounting is unit-tested without a real AWS client or SSH. Reload failures are now tracked in a separate reloadFailedInstances list, printed with the same manual-fallback message as the single-instance path, and reflected in the summary output (Reloaded: N/M instances) and audit log (reload_failed_count).

Test plan

• go build ./...
• go vet ./cmd/...
• go test ./cmd/... — new tests: TestExtendJobArrayInstances_ReloadFailureIsTrackedNotDiscarded, TestExtendJobArrayInstances_TagFailureSkipsReload, TestExtendJobArrayInstances_AllSucceed
• golangci-lint run ./cmd/... — no new findings in touched files (pre-existing findings in unrelated files unchanged)
• CHANGELOG.md updated under [Unreleased]

Fixes #512

[codecov]

Codecov Report

❌ Patch coverage is 44.82759% with 16 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
cmd/extend.go	44.82%	16 Missing ⚠️

📢 Thoughts on this report? Let us know!