Fix: Three defects in the buildings toolkit (B236, B238, B242)

[ilayfalach]

Fixes for 3 defects from the unit-test expansion log: B236, B238, B242.

The work is done and committed on the local branch fix/groupU-more, cut from tests/batch33-coverage90. Nothing is pushed yet — these branches can only land after the test-expansion stack (#1067, #1083, #1084, #1085), because every fix is verified against the xfail(strict=True) test that pins it.

What each commit says

f411f879 fix: three defects in the buildings toolkit (B236, B238, B242)

- B236 `getBuildingHeightFromRasterTopographyToolkit` accepts
       topographyDataSource, documents it as "the name of the datasource in
       the topography toolkit", and then never forwards it -- so the argument
       had no effect and even a datasource that does not exist was ignored.
- B238 the module's `try: import FreeCAD; import Part; import Mesh` left
       FreeCAD bound when only Part or Mesh was missing, so the method's own
       "FreeCAD not found. Install before using this function" guard never
       fired and a partial install surfaced later as a bare NameError on Part.
       All three names are now unbound together.
- B242 the lambda cache descriptor carried `"bounds": data.total_bounds`, a
       numpy.ndarray. A document stored with an array-valued field cannot be
       matched again, so the second identical call did not hit the cache and
       poisoned every later lookup, overwrite included. Stored as a list.

Unit layer: 5584 passed, 375 xfailed, 39 failed -- no regressions.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Verification

Full unit layer run after every commit, with the failure list diffed against a baseline captured before any edit: no regressions. The 39 baseline failures are environmental (no OpenFOAM, no current hermes, no gdal locally).