Expose missing GUI scoring tools

[panDing19]

Problem

Several GUI tasks rely on app-direct GUI tools for scoring or expected actions, but the task metadata does not expose those tools. Agents running against the declared tool set cannot call tools that the benchmark expects them to use.

Affected tasks

• TGUI01 through TGUI06
• TGUI23
• TGUI44
• TGUI49
• TGUI50

Fix

• Add the missing GUI tool specs and endpoints to the affected task metadata.
• Cover My Expenses, Loop Habit, Clock, Gmail Clone, Contacts, Fossify Calendar, and TestMall app-direct tools where required.
• Add a regression test that checks the affected tasks expose both tools and tool_endpoints entries for the required scoring tools.

Tests

• uv run pytest tests/test_gui_missing_tool_exposure.py -q