Live chamfer and fillet tests aim at the wrong faces of their test block #199

The live tests in tests/live/test_live_sw_add_chamfer_tool.py use fixed points for the faces and edges they chamfer and fillet. Those points assume the test block's height runs along Z. It actually runs along Y: the block is sketched on the Top Plane and extruded 15 mm, and its measured bounding box is min [0, 0, -30], max [30, 15, 0] (mm). So the tests have not been checking what they say they check.

What each constant actually hits

Evidence: a 0.5 mm add_chamfer on _TOP_FACE removed 11.08 mm³. That's what a side face gives: two 30 mm edges and two 15 mm edges come to about 11.25 mm³. The face with the hole should give about 17.3 mm³ (the hole rim plus four 30 mm edges). Using a point on the right face, face:0.005,0,-0.025, removed 17.12 mm³, which is 99% of that expectation. A visual check confirmed the chamfer was on a short side.

Impact: test_add_chamfer_tool_whole_top_face requires more than 13.8 mm³, so it would fail against this geometry. The mixed face-and-edge test and the chamfer-plus-fillet test only check that some volume changed, so they pass while working on the wrong faces. The module docstring's history (the "3.75 mm³ wrong edge" finding) rests on the same wrong axis assumption, so it needs rechecking.

Proposed fix:

• Work each face and edge point out from the block's measured bounding box, not from hard-coded coordinates. The API-lookup demo in tests/live/test_live_sw_api_lookup.py already does this.
• Correct the edge constant to a real vertical edge, e.g. (0, 0.0075, 0).
• Give every test a volume window from the geometry, so a chamfer on the wrong face fails.
• Update the docstring.
• Run the suite live to confirm.

The add_chamfer and add_fillet tools themselves behave correctly: they act on the face the given point lies on.