Add guarded input target assertions

[zengfanfan]

Summary

Adds optional shared foreground target assertions to input tools that send foreground input:

• Click
• Type
• Scroll
• Move
• Shortcut

The new optional parameters are expected_window_title and expected_process. When supplied, Windows-MCP validates the current foreground root window before sending input and fails before the action if the target does not match.

Motivation

Agents often observe a target window and then send input in a later tool call. Today the input call trusts whichever window is foreground at execution time. These guards give callers an opt-in fail-closed check for wrong-target input.

Issue: #314

Public API and compatibility

Existing unguarded calls remain unchanged. Guarded calls can opt in like this:

Click(loc=[x, y], expected_window_title="Notepad", expected_process="notepad.exe")
Type(text="hello", loc=[x, y], expected_window_title="Notepad")
Scroll(loc=[x, y], expected_process="notepad.exe")
Move(loc=[x, y], expected_window_title="Notepad")
Shortcut(shortcut="ctrl+s", expected_process="notepad.exe")

Matching semantics:

• expected_window_title: case-insensitive substring match against the foreground root title;
• expected_process: case-insensitive exact executable basename match;
• if both are supplied, both must match;
• mismatch fails before low-level input.

Implementation overview

• Desktop.assert_foreground_target(...) centralizes foreground title/process validation.
• Input tools call the shared assertion before their low-level action.
• README documents the optional guard parameters.
• Tests cover legacy behavior, guard success, guard failure-before-input, both-guard forwarding, missing foreground, and title/process mismatch.

Testing

Focused checks passed:

uv run pytest tests/test_guarded_input_tools.py tests/test_foreground_target_assertions.py
uv run ruff check src/windows_mcp/desktop/service.py src/windows_mcp/tools/input.py tests/test_guarded_input_tools.py tests/test_foreground_target_assertions.py

Focused pytest result:

16 passed

Full repository gates were also run. They still show pre-existing baseline failures unrelated to this PR:

uv run pytest
# 334 passed, 3 failed in tests/test_iserror_compliance.py due missing shell/clipboard/registry exported tool functions

uv run ruff check .
# 64 pre-existing lint errors

Notes

This PR is intentionally separate from deterministic drag PR #313. It does not include local planning documents, private consumer details, or unrelated cleanup.