Improve handling of blocksize_parameter

[FynnBe]

If a high blocksize_parameter is given to predict it fails if the actual sample is smaller than the deduced block shape.
For simple blocksize_parameter values (integers) a smaller value could be chosen automatically to avoid this.

[falibabaei]

Hi Fynn,

Would it be possible to pad the input to avoid this issue, rather than choosing a smaller value for the blocksize_parameter?

[FynnBe]

Would it be possible to pad the input to avoid this issue, rather than choosing a smaller value for the blocksize_parameter?

yes, to some extend. My plan is actually to do both.
As we wouldn't want to pad more than we need to get a valid shape.
Otherwise one might accidentally pad to a multiple of the image size, etc..