[CK_TILE] Add save_matrix_txt() and extract HostTensor I/O to free functions (#6807)
## Summary
- Extract `loadtxt`, `savetxt`, and `save_matrix_txt` from `HostTensor`
member functions into standalone free functions in `host_tensor_io.hpp`
(Single Responsibility Principle)
- Add `save_matrix_txt()` for writing 2D tensors to space-separated text
files with configurable output limit (default 256x256, pass 0 to dump
all)
- Supports float, int, and int8_t output formats via a `dtype` parameter
- Validate dtype early and throw on unsupported values in all three
functions
- Update callers in `15_fused_moe/main.cpp` to use free function syntax
0df3523ef1