Improve OpenCode identity removal accuracy (#87)

This pull request improves the accuracy of OpenCode identity removal from system prompts. The key changes include:

- Changed `OPENCODE_IDENTITY` constant from a full sentence to `OPENCODE_IDENTITY_PREFIX` containing just "You are OpenCode" for more precise matching
- Updated `sanitizeSystemText()` to remove entire paragraphs containing the OpenCode identity prefix rather than attempting inline removal
- Refactored `prefixToolNames()` to accept a parsed object instead of a JSON string, eliminating redundant parsing logic in `rewriteRequestBody()`
- Added IDE configuration files for Biome formatting and a project dictionary
- Updated all related tests to use the new prefix-based approach

The changeset indicates this is a patch-level change focused on more accurate OpenCode identity removal.

<!-- CURSOR_SUMMARY -->
---

> [!NOTE]
> **Medium Risk**
> Changes request-body rewriting and system-prompt sanitization logic, which could affect how prompts/tools are sent to Anthropic and potentially alter downstream behavior if the new matching is too broad or too narrow.
> 
> **Overview**
> Improves system-prompt debranding by replacing the exact `OPENCODE_IDENTITY` match with an `OPENCODE_IDENTITY_PREFIX` check and updating `sanitizeSystemText()` to drop *any paragraph* containing that prefix (instead of trying inline removal).
> 
> Refactors `prefixToolNames()` to operate on an already-parsed request object (removing its internal JSON parse/invalid-JSON fallback) and updates `rewriteRequestBody()` to reuse it; tests are adjusted accordingly. Adds a patch changeset and some IDE config files.
> 
> <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit ebc03d9e8d07dc17b1346356cf537735482e0166. Configure [here](https://www.cursor.com/dashboard/bugbot).</sup>
<!-- /CURSOR_SUMMARY -->