Depending on the engine used linalg.qr returns full or reduced QR decomposition

When using native engine (enabled by default on my system) QR decomposition seems so be returning reduced matrices.
When switching to kotlin engine via `Multik.setEngine(KEEngineType)` it returns full Q and R matrices.

Start by locating the linalg.qr implementations used by the native engine and by Multik.setEngine(KEEngineType), then compare their returned matrix shapes. Check existing QR-related tests or add a regression test that exercises both engines and documents the expected full or reduced decomposition behavior.

Written by the indexing model from the issue text.