SpatialPad on the inferer when the to-pad shape is the same as the tensor

[virginiafdez]

In the Latent inferers, in some cases, if the latent tensor shape is the same as the one you have to pad to (because the users populated the fields ldm_latent_shape and vae_latent_shape), the code fails because the latent tensor's got a track_meta flag that tries to calculate the inverse transform when calling the resizer and it fails with an error.

A check, before padding, that verifies that padding is necessary (ldm_latent_shape != latent.shape) will overcome this problem.