fix: Partial acceptance of records in ProduceSync

[grobinson-grafana]

What this PR does / why we need it:

This commit fixes situations where some records in a batch can be accepted, but not others. For example, consider a limit of 1KB, and records of size 512B, 513B and 256B. In this case, records 1 and 3 will be accepted, and 2 will be rejected. In most cases, this results in retries, and duplication of data. This change means that the entire batch is rejected if any records exceed the limit.

Same fix in Mimir: grafana/mimir#15227

Which issue(s) this PR fixes:
Fixes #

Special notes for your reviewer:

Checklist

• Reviewed the CONTRIBUTING.md guide (required)
• Documentation added
• Tests updated
• Title matches the required conventional commits format, see here
• Changes that require user attention or interaction to upgrade are documented in docs/sources/setup/upgrade/_index.md
• If the change is deprecating or removing a configuration option, update the deprecated-config.yaml and deleted-config.yaml files respectively in the tools/deprecated-config-checker directory. Example PR

---

Note

Medium Risk
Changes producer backpressure behavior by failing an entire batch when the buffered-bytes limit would be exceeded, which could affect ingestion under load. Logic is localized and covered by a new unit test, keeping overall risk moderate.

Overview
Producer.ProduceSync now enforces the buffered-bytes limit atomically per batch: it sums record sizes up-front and, if the batch would exceed maxBufferedBytes, it fails all records with kgo.ErrMaxBuffered instead of partially accepting some.

Adds a test ensuring oversized batches are fully rejected and that the same records succeed when produced individually.

^{Reviewed by Cursor Bugbot for commit ac8701b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[ashwanthgoli]

lgtm