Bug
In crates/apr-cli/src/commands/slice.rs, the output_slice_result() function hardcodes the JSON shape field as [count] (line 193), discarding the actual tensor shape.
The slice_gguf() function retrieves the real shape (line 112: let (data, shape) = get_gguf_tensor_f32(...)) but then explicitly discards it (line 134: let _ = shape;).
Impact
When slicing a 2D tensor [768, 3072] and extracting elements 0..10, JSON output shows:
instead of:
{"shape": [768, 3072], "slice_count": 10, ...}
Fix
Pass the actual tensor shape through to output_slice_result() and include it in JSON output alongside slice count.