Motivation & Context

HeadTailBuffer can silently drop and reorder multi-byte UTF-8 output when a complete rune no longer fits the remaining head bytes but later smaller runes still do. At the exact overall byte cap, the previous logic could evict a tail rune and still report that the output was not truncated. This affects the bounded stdout and stderr capture shared by the local and Docker shell executors.

Description & Review Guide

• What are the major changes? Seal the head after the first rune spills to the tail, defer whole-rune tail eviction until total input exceeds the overall cap, and add exact-cap and overflow regressions.
• What is the impact of these changes? Inputs that fit the configured byte cap now round-trip in order without silent loss, while overflow remains bounded and retains only complete UTF-8 runes.
• What do you want reviewers to focus on? The one-way transition from head to tail and the eviction boundary when total input first exceeds the cap.

Related Issue

Fixes #7112

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.