Remove isContextOverflowError check from sanitizeUserFacingText to prevent false positives when the assistant mentions phrases like "context overflow" in normal conversation.

Real context overflow errors are already handled by formatAssistantErrorText, which properly checks msg.stopReason === "error" before applying error transformations.

Fixes #7483

Greptile Overview

Greptile Summary

This PR removes the isContextOverflowError() rewrite from sanitizeUserFacingText() to avoid false positives when normal assistant text mentions phrases like “context overflow”. Actual LLM error strings are still rewritten by formatAssistantErrorText(), which gates error-specific formatting on msg.stopReason === "error" (or an explicit errorMessage).

The change is localized to src/agents/pi-embedded-helpers/errors.ts and narrows the scope of “context overflow” handling to the error-formatting path rather than the general user-facing text sanitization path.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk.
• The change is a small, targeted removal of an over-broad heuristic that could misclassify ordinary assistant text. Context overflow handling remains in the dedicated error formatting path, so the user-facing error UX should remain intact while reducing false positives.
• No files require special attention