Report actual per-entry idle time for autoclaimed background jobs

[beranekio]

Summary

ClaimBackgroundJobs currently sets idle_ms on autoclaimed BackgroundJob and PendingBackgroundJob entries to autoclaim_min_idle_ms — a lower bound, not the entry's true PEL idle duration. PR #42 documented this as minimum-threshold semantics, but downstream workers (for example duihua-background-worker) may need the actual idle time for stale-job detection, retry/backoff, and distinguishing freshly claimed work from long-abandoned PEL entries.

Redis XAUTOCLAIM only returns stream entries that exceeded the threshold; it does not include per-entry idle metadata in the reply.

Problem

An entry idle for minutes but reclaimed with autoclaim_min_idle_ms = 50 is exposed as idle_ms = 50. Workers treating idle_ms as exact idle duration will substantially understate abandonment time and may schedule retries as live work instead of interrupted/stale handling.

Related review: Codex on PR #42 — #42 (comment) (approx.)

Proposed direction

Extend the proto/API to expose both values explicitly, for example:

Option A — add idle_ms_actual (preferred)

Keep existing idle_ms as documented minimum threshold (autoclaim_min_idle_ms) for backward compatibility, and add:

message BackgroundJob {
  // ...
  optional uint64 idle_ms_actual = 6; // true PEL idle when autoclaimed
}

message PendingBackgroundJob {
  // ...
  optional uint64 idle_ms_actual = 5;
}

Populate idle_ms_actual from Redis XPENDING (or equivalent) for autoclaimed entries in the claim batch.

Option B — repurpose idle_ms

Change idle_ms to mean actual idle time and add idle_ms_min for the reclaim threshold. Clearer semantics but breaking for consumers that already read PR #42's minimum-threshold docs.

Implementation notes

• Fetch per-entry idle in the autoclaim hydration path (jobs_from_stream_ids / claim_jobs) before or alongside XAUTOCLAIM output mapping.
• XPENDING returns last_delivered_ms per stream ID; verify whether post-XAUTOCLAIM idle is still useful or whether pre-claim lookup is required.
• Batch XPENDING where possible to avoid N+1 round-trips on large autoclaim batches.
• Propagate fields through Rust core/client, server conversion, and Go SDK.
• Update tests in crates/core/tests/pending_job_metadata.rs to assert idle_ms_actual exceeds the minimum threshold when entries have aged.

Acceptance criteria

• Proto exposes actual autoclaim idle separately from the reclaim threshold (or documents and implements a single field with guaranteed accurate semantics)
• Rust and Go SDKs regenerated and committed
• Integration test proves an entry idle longer than autoclaim_min_idle_ms reports idle_ms_actual above the threshold
• README documents both fields and intended worker usage
• Downstream worker can use actual idle for stale/autoclaim retry policy without underestimating abandonment

Related

• Include claim metadata on PendingBackgroundJob #41 — original PendingBackgroundJob claim metadata work
• PR Include claim metadata on PendingBackgroundJob #42 — minimum-threshold documentation interim fix
• Preserve autoclaim metadata when scheduling pending hydration retries duihua-background-worker#13 — worker stale/autoclaim retry behavior