Optimize IPCCoupler with Taichi kernels

- Add IPCTransformData and IPCCouplingData classes following Genesis pattern
- Pre-allocate all arrays as Taichi fields, eliminate runtime allocations
- Refactor _set_genesis_transforms_from_ipc with parallel kernel processing
- Add initialization and filtering kernels for parallel processing
- Rename _compute_external_wrench_kernel to _compute_external_force_kernel

Author: Roushelfy