Add: BGEMM example for tensormap_and_ringbuffer runtime

[chenshengxin2026]

Add BGEMM (Batched General Matrix Multiplication) example demonstrating
Cube (AIC) and Vector (AIV) core cooperation with tile-based computation
using the tensormap_and_ringbuffer runtime.

Add examples/tensormap_and_ringbuffer/bgemm/

• Task dependencies resolved automatically via TensorMap overlap detection
• Supports 4x4x4 grid with 64x64 tiles (256x256 matrices, batch=2)

Example includes:

• golden.py: Test specification with tile-first 5D memory layout
• kernel_config.py: Runtime and kernel configuration
• kernels/aic/kernel_gemm_tile.cpp: Cube core matmul kernel (TMATMUL)
• kernels/aiv/kernel_tile_add.cpp: Vector core accumulation kernel (TADD)
• kernels/orchestration/bgemm_orch.cpp: Task graph builder

Closes #84