fix(chat): sync live-stream fence strip regex with backend TOOL_TAGS

[michaelxer]

Summary

The browser-side EXEC_FENCE_RE regex in chatRenderer.js only covered 6 tool names (web_search, read_file, write_file, create_document, edit_document, update_document). The backend strips all TOOL_TAGS from persisted round_texts via tool_parsing.py, but the live streaming path uses the narrower frontend regex. This mismatch causes executed fenced tool calls for email, cookbook, MCP, and other tools to remain visible in the live chat view until the session is reloaded.

This PR replaces the hardcoded EXEC_FENCE_RE with an EXEC_TOOL_TAGS array that mirrors src/agent_tools/__init__.py TOOL_TAGS (excluding bash, python, ls which are valid fenced-code language identifiers), and builds the regex from it.

Target branch

• This PR targets dev, not main.

Linked Issue

Fixes #3993

Type of Change

• Bug fix (non-breaking — fixes a confirmed issue)

Checklist

• I searched open issues and open PRs — this is not a duplicate.
• This PR targets dev
• My changes are limited to the scope described above.
• I actually ran the app and verified the change works end-to-end.

How to Test

1. Start Odysseus with a local fenced-tool model (e.g. Ollama).
2. Enable an email tool (e.g. list_emails) via agent tool config.
3. Ask the agent to list emails so the model emits a fenced list_emails call.
4. Watch the live streaming bubble during and after tool execution.
5. Before fix: the raw fenced block remains visible until session reload.
6. After fix: the fence is stripped from the live view immediately.

Regex verification:

• EXEC_FENCE_RE.test("```list_emails\n{\"max_results\":10}\n```") -> true
• EXEC_FENCE_RE.test("```python\nprint(\"hello\")\n```") -> false

Visual / UI changes

• No visible UI change — this only affects text stripping logic in stripToolBlocks(). No new CSS, colors, fonts, or component patterns.

[michaelxer]

CI is all green, ready for review when you have a moment.

[michaelxer]

Closing to keep the PR queue tidy. The fix is still valid if anyone wants to pick it up -- the frontend EXEC_FENCE_RE regex only covers 6 tool names while the backend strips all TOOL_TAGS, leaving email/cookcook/MCP tool calls visible during live streaming.