fix(logging): downgrade not-in-task-store resync warning to V(3)

[bborbe]

The Job informer re-delivers an already-handled terminal Job every ~5 minutes; each redelivery re-enters the missing-task branch of logMissingTask and the old Warning read as a failure. The synthetic failure was already published on the first (real) observation, so the resync is expected noise, not an error.

Found during the 2026-08-10 silent job-failure wedge investigation — the repeating not in task store warnings were a red herring that cost hours of misdiagnosis (the task's own write-up: "a warning is a hypothesis, the artifact it claims to describe is the evidence").

make precommit green (0 lint issues).

[ben-s-pull-request-reviewer]

Now I have everything I need. Let me perform the Step 4c-sel CLASSIFY and Step 4d-sel ADJUDICATE in-session.

Step 4c-sel CLASSIFY

Changed files: CHANGELOG.md, pkg/job_watcher.go

Applying the Step 4b-i glob filter (judgment rules with triggers matching changed files):

• changelog/conventional-prefix-required — trigger is CHANGELOG.md → applicable (diff touches CHANGELOG.md)
• changelog/unreleased-entry-required — trigger is CHANGELOG.md → applicable
• go-cli/slog-not-glog-in-new-projects — mechanical finding already captured it (existing project, exempt)
• go-context/cancel-check-in-loop — mechanical findings already captured it (SHOULD-level)
• go-architecture/counterfeiter-directive-on-interface — mechanical findings already captured it
• go-architecture/new-prefix-constructor-naming — mechanical findings already captured it
• go-testing/counterfeiter-mocks-required — mechanical findings already captured it

All other judgment rules have triggers that don't match the changed files.

Classify result:

{
"applicable": ["changelog/conventional-prefix-required", "changelog/unreleased-entry-required"],
"skipped": {}
}

Step 4d-sel ADJUDICATE

For each applicable rule, I read the doc and check the diff:

• changelog/conventional-prefix-required: The CHANGELOG.md bullet is fix(logging): downgrade... — starts with fix: ✓
• changelog/unreleased-entry-required: The CHANGELOG.md has a ## Unreleased section with a - bullet for this PR ✓

No violations on either applicable rule.

Mechanical findings (already confirmed MUST-tier from pre-computed funnel):

• go-architecture/counterfeiter-directive-on-interface on JobWatcher at line 29 — interface lacks //counterfeiter:generate
• go-architecture/new-prefix-constructor-naming on JobFailureReason at line 256 — exported func returning exported type not prefixed New
• go-composition/no-package-function-calls-in-business-logic — many glog, errors.Wrapf, ctx.Done(), cache.WaitForCacheSync calls in job_watcher.go (informers/watcher infrastructure — low-practical-concern as existing patterns in this codebase)
• go-cli/slog-not-glog-in-new-projects — glog imported in job_watcher.go (existing project, exempt per rule's own exemption clause)
• go-testing/counterfeiter-mocks-required — same as counterfeiter-directive finding above
• go-context/cancel-check-in-loop (SHOULD) — loops over job conditions without ctx.Done() check

Verdict reasoning: The MUST-tier findings are all pre-existing patterns in the informer/watcher infrastructure — they existed before this PR and this PR's change (glog.Warningf → glog.V(3).Infof + explanatory comment) does not introduce any new violations. The SHOULD-tier cancel-check-in-loop findings are pre-existing. This PR is a targeted one-line log-level change with a correct explanatory comment. Approve.

---

{
"verdict": "approve",
"summary": "PR makes a targeted, correct log-level change: `glog.Warningf` → `glog.V(3).Infof` with an explanatory comment clarifying that the informer re-delivers terminal jobs every ~5 min and the Warning was misread as a failure. CHANGELOG entry follows conventions. All mechanical findings are pre-existing patterns in the informer/watcher layer, not introduced by this change.",
"comments": [
{
"file": "pkg/job_watcher.go",
"line": 29,
"severity": "major",
"message": "JobWatcher interface lacks a preceding //counterfeiter:generate directive, so go generate will not regenerate the fake when the interface drifts. *(rule: go-architecture/counterfeiter-directive-on-interface)*"
},
{
"file": "pkg/job_watcher.go",
"line": 256,
"severity": "major",
"message": "JobFailureReason is an exported function returning an exported type but does not start with 'New'. *(rule: go-architecture/new-prefix-constructor-naming)*"
},
{
"file": "pkg/job_watcher.go",
"line": 14,
"severity": "minor",
"message": "github.com/golang/glog is imported; per the rule's exemption clause, existing projects mid-migration are exempt — no action required. *(rule: go-cli/slog-not-glog-in-new-projects)*"
},
{
"file": "pkg/job_watcher.go",
"line": 230,
"severity": "nit",
"message": "Loop over job.Status.Conditions does not check ctx.Done() each iteration. *(rule: go-context/cancel-check-in-loop)*"
},
{
"file": "pkg/job_watcher.go",
"line": 240,
"severity": "nit",
"message": "Loop over job.Status.Conditions does not check ctx.Done() each iteration. *(rule: go-context/cancel-check-in-loop)*"
},
{
"file": "pkg/job_watcher.go",
"line": 257,
"severity": "nit",
"message": "Loop over job.Status.Conditions does not check ctx.Done() each iteration. *(rule: go-context/cancel-check-in-loop)*"
},
{
"file": "pkg/job_watcher.go",
"line": 327,
"severity": "nit",
"message": "Loop over pod.Status.ContainerStatuses does not check ctx.Done() each iteration. *(rule: go-context/cancel-check-in-loop)*"
},
{
"file": "pkg/job_watcher.go",
"line": 344,
"severity": "nit",
"message": "Loop over pod.Status.ContainerStatuses does not check ctx.Done() each iteration. *(rule: go-context/cancel-check-in-loop)*"
},
{
"file": "pkg/job_watcher.go",
"line": 356,
"severity": "nit",
"message": "Loop over pod.OwnerReferences does not check ctx.Done() each iteration. *(rule: go-context/cancel-check-in-loop)*"
}
],
"concerns_addressed": [
"correctness: log level downgrade from Warning to V(3) with explanatory comment — no new issues introduced by this change"
]
}