fix(core): guard non-reducing compaction

[cwj2001]

Issue for this PR

Fixes #27924

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Stops overflow recovery when compaction does not make progress.

After the recovery summary is added, the runner re-estimates the request. It only retries the provider if the compacted request is smaller than the original request and within the model context budget. Otherwise it persists the original overflow error and exits instead of retrying into another overflow cycle.

How did you verify your code works?

From packages/core:

• bun test test/session-runner.test.ts --test-name-pattern "overflow"
• bun typecheck

Screenshots / recordings

Not a UI change.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Based on my search, I found a potentially related PR:

Related PR:

• fix(core): handle aborted overflow compaction #34029 - fix(core): handle aborted overflow compaction
  • URL: fix(core): handle aborted overflow compaction #34029
  • This PR addresses overflow compaction handling in the core package, which is closely related to PR fix(core): guard non-reducing compaction #34261's focus on guarding against non-reducing compaction and preventing overflow cycles.

The other PRs found (#33667, #18941) are related to overflow and compaction in general but appear to address different aspects of the issue.

[cwj2001]

#34029 and this PR cover different failure modes.

#34029 handles aborts/errors while generating the overflow recovery summary.

This PR handles the case where recovery summary generation succeeds, but the rebuilt request is still over budget or is not smaller than the original request. In that case we persist the original overflow error instead of retrying into another overflow cycle.