Preserve pre-tool hook feedback in model history

[zikajk]

Summary

• Deliver successful preToolCall.additionalContext with the matching tool result, rather than showing warnings only in the UI. Existing updatedInput behavior is unchanged.
• Append feedback after post hooks so output replacement cannot erase it; ignore blank/failed-hook context and avoid duplicate denial messages.
• Keep four focused regression cases and concise hook documentation. Plugin merging and brepl changes are separate.

Verification

• 68 tests, 344 assertions passed across tool calls, chat lifecycle, and hooks.
• Focused Clojure lint, editor diagnostics, and whitespace checks passed.

🤖 Generated with ECA (openai/gpt-6-astra)

[zikajk]

Closing as out of scope for the brepl integration. Pre-tool hooks will continue to adjust or reject input; post-tool hooks will report repairs and evaluation results. The existing updatedInput behavior is sufficient, so this general pre-hook context-delivery extension is not needed. PR #592 is separate and unchanged.