[ankur22]

Follow-up to #6171 (now merged), rebased directly onto master. Addresses two review threads deferred from that PR. All commits signed.

What?

Two fixes deferred from #6171's review:

1. Drain the loki hook on shutdown (internal/log/loki.go). lokiHook.Listen's ctx.Done() branch pushed only the entries it had already consumed into its buffer and never drained what was still queued in the channel, so a shutdown could lose a tail of logs. It now drains the channel before the final push. This is shared code — it also fixes --out loki's final-batch loss — and the loss was made observable by Stream --local-execution logs to Grafana Cloud #6171's cloud-log Drain() path. Split into a pure-refactor commit (extract the per-entry buffering into an appendEntry helper) and the fix itself.

2. Report dropped cloud logs (internal/log/cloud/cloud.go). The cloud log Pusher.Fire counted entries it dropped when the front buffer was full, but nothing read that counter, so a full buffer lost logs silently. It now emits a synthetic k6 dropped N log messages… warning at shutdown, mirroring how --out loki reports its own limit drops. The notice is pushed via a new log.NoticePusher capability on the loki hook, which sends it as a standalone request exempt from the hook's per-batch limit (a full final batch would otherwise swallow it), and it is pushed after the final flush so it carries the newest timestamp in the warning stream it shares with the hook's own limit-drop message — avoiding an out-of-order write that Loki rejects with unordered_writes=false.

Three atomic commits, each individually buildable + lintable + green under make check.

Why?

Both were flagged in #6171's review as pre-existing gaps that its Drain() path exposed, but that belonged in a dedicated follow-up rather than widening that PR:

• The shutdown tail-loss lives in the shared loki hook (also used by --out loki), so fixing it there benefits both paths and was kept out of the cloud-logs PR.
• The write-only dropped counter only becomes reliably reportable once the drain is fixed — otherwise the "dropped N" notice could itself be lost on shutdown, which is why it is sequenced after the drain fix here.

Checklist

• I have performed a self-review of my code.
• I have commented on my code, particularly in hard-to-understand areas.
• I have added tests for my changes.
• I have run linter and tests locally (make check) and all pass.

Related PR(s)/Issue(s)

• Follows Stream --local-execution logs to Grafana Cloud #6171 (cloud log streaming for k6 cloud run --local-execution), now merged; addresses two of its deferred review threads.
• Drain tail-loss discussion: Stream --local-execution logs to Grafana Cloud #6171 (comment)

The base branch was changed.