BUG: Cache validation does not check latent size for SD/SDXL (sd3 branch)

The new cache validation in the `sd3` branch only checks the key name. When multiresolution is disabled (like with sd and sdxl training), the key is `latents` and the size of the tensor is not checked. If the training resolution or bucketing changes, the cache should not pass validation.

https://github.com/kohya-ss/sd-scripts/blob/123474d784096a2a3063d1ff91ef1957b41ccc72/library/strategy_base.py#L446-L451

Start in library/strategy_base.py at lines 446-451 and inspect how cache validation handles the `latents` key when multiresolution is disabled. Trace the existing validation path for SD and SDXL training, then verify that changing the training resolution or bucketing causes the cache to fail validation.

Written by the indexing model from the issue text.