resize-bilinear: guard indirection buffer and packed weights size against overflow

[destro4evr-rgb]

xnn_reshape_resize_bilinear2d_nhwc_* and xnn_reshape_resize_bilinear2d_nchw_* computed the indirection buffer size and packed-weights size with raw multiplication:

sizeof(void*) * (output_height * output_width * 4)
(output_height * output_width * 2) << log2_weight_element_size

On 32-bit targets (WebAssembly, ARM32) where size_t is 32 bits, the product overflows when output_height * output_width >= 2^28 (e.g. 16384 × 16384), wrapping the computed size to zero. xnn_reallocate_memory then returns a zero-size allocation, and xnn_indirection_init_resize_bilinear2d_* subsequently writes 4 * output_height * output_width pointer entries into this undersized buffer — a heap out-of-bounds write.

Replace the raw multiplications with chained xnn_safe_mul calls that return xnn_status_out_of_memory on overflow, matching the pattern used in other reshape paths.