fix: status no longer presents an unread TTL as a resolved "none"

[scttfrdmn]

Summary

Three related defects in spawn status/spored status, all from the same field incident (#508):

1. TTL: none was rendered from data that was never read. When spored couldn't read its own tags (the CRITICAL: launch --iam-policy-file omits the spored baseline policy — TTL, --on-complete and --pre-stop all silently dead (the #406 fix covers 2 of 3 role paths) #502 IAM gap, or any ec2:DescribeTags denial), the load failure was silently swallowed into zero-value config defaults, and status printed TTL: none — instance will not auto-terminate as a definite claim. Added Config.ConfigLoadError (threaded from pkg/provider/ec2.go's tag-load failure) so cmd/spored/main.go's status renderer can print TTL: UNKNOWN — could not read config (<error>) instead when the config was never actually resolved.
2. The two lifecycle lines were never reconciled. That on-instance "none"/"UNKNOWN" line could sit directly above spawn status's own Termination deadline: <t> line — read from the same tags but with the caller's credentials, which usually succeed even when the instance role can't. Added ttlReconciliationNotice in cmd/status.go: when the on-instance view is unresolved AND a tag-based deadline exists, it prints an explicit "Lifecycle mismatch" notice naming both facts and what they mean together (the deadline exists, nothing on the instance can see it, so nothing on the instance will enforce it) — the sentence the reporter had to assemble by hand.
3. Started/Elapsed described the CLI/agent invocation's own age, not the instance's. startTime := time.Now().Add(-uptime) was the fallback whenever spawn:launch-time couldn't be read — exactly defect 1's failure mode — so a 7h39m-old instance reported Elapsed: 0s. Added a middle fallback tier: identity.PendingTime, EC2's own record of the instance's launch request time from the instance identity document via IMDS — no IAM permission required, so it survives a tag-read failure. Extracted resolveLaunchTime as a pure function and label the source inline whenever it isn't the authoritative tag.

Also flagged in the issue but not addressed here (out of scope / not reproduced): the observation that the CPU/network panel numbers looked byte-identical across 7 instances — that's a separate, unconfirmed claim about the metrics-sampling path, not the TTL/Elapsed rendering this issue is about; happy to investigate separately if it recurs.

Test plan

• go build ./..., go vet ./..., gofmt -l . (clean)
• go test ./cmd/... ./cmd/spored/... ./pkg/provider/... ./pkg/agent/...
  • New: TestResolveLaunchTime (cmd/spored) — verified the IMDS-fallback case fails against a pre-fix version that skips straight to the agent's own start time, matching the exact status: prints TTL: none — instance will not auto-terminate and a Termination deadline in the same screen, from two unreconciled sources — and Started/Elapsed describe the CLI invocation, not the instance (Elapsed: 0s at 7h39m) #508 symptom (Elapsed: 0s on a 7h-old instance)
  • New: TestTTLReconciliationNotice (cmd) — five cases covering both unresolved-TTL renderings (none and UNKNOWN), the past-due variant, and the two "nothing to reconcile" cases (no tag deadline; on-instance already resolved)
• golangci-lint run ./... — no new findings in touched files
• CHANGELOG.md updated under [Unreleased]

Fixes #508

[codecov]

Codecov Report

❌ Patch coverage is 66.66667% with 12 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
cmd/spored/main.go	42.85%	8 Missing ⚠️
cmd/status.go	90.00%	2 Missing ⚠️
pkg/provider/ec2.go	0.00%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!