Allow quantize to only copy tensors, other improvements

[KerfuffleV2]

You can now specify copy as the type to the quantize tool to only copy tensors, never quantize. I forget where I saw it, but it was recently mentioned that something like could be used to convert/repackage from stuff like GGUF V1 to GGUF V2.

This also improves the logic in the actual quantize function (mainly for k-quants) to be a bit smarter when requantizing to the same format. The current logic skips quantizing when quantize_type == tensor->type. However, this means the k-quants logic for stuff like counting numbers of tensors to decide whether to use more bits doesn't run. The result is if you quantize to q4_k_m and then requantize to q4_k_m stuff like requantizing q6_k tensors to q4_k will occur. Not super likely, but I recall reading a pull recently where someone was confused by behavior like that and making it work a little more intuitively is pretty easy. (Of course, if the k-quants strategy changes then you're still going to have a bad time but this change is at least as good as the status quo.)

[ghost]

Related: #2821 (comment)