quantize: Handle user-defined pruning of whole layers (blocks)

[EAddario]

This PR adds the ability to prune all tensors from user-defined layers (blocks) by providing a comma-separated list in the --prune-layers command line option. It will renumber remaining layers to avoid gaps in the sequence, update the relevant model metadata and, if an imatrix is available, it will use the correct importance score vector.

Pruning is restricted to repeating layers only (i.e. blk.n, blk.n+1, etc.) and will not affect single tensors like output, token_embd, etc.

This option can be used alongside --tensor-type to perform tensor/layer-wise quantization on selected tensor types, whilst at the same time pruning others. For example:

llama-quantize --tensor-type attn=q6_k --prune-layers 3,7,11 --imatrix imatrix.dat model-f32.gguf model-q4_k_m.gguf q4_k_m

It was inspired partly by ShortGPT: Layers in Large Language Models are More Redundant Than You Expect and partly as the next logical step from #12511. It could be used alongside #12718 to guide the layer selection.

Opening a draft PR for now until split tensor testing is completed, but feedback and suggestions are encouraged in the meantime.

[EAddario]

Finally finished testing several combinations with split files and it seems to work 100%

Examples of pruned models here and here

[EAddario]

Apologies for shotgun approach @slaren / @CISC / @ggerganov, not sure what the proper process to request a review is. Happy to close or move to draft if it's not suitable for merging

[EAddario]

Thank you @CISC!

[CISC]

nits :)

[EAddario]

@CISC, much appreciated for spending time on this. Fixed the issues and ran different combinations of tests (split files in -> single out; single file in -> split out; split in -> split out; single in -> single out) pruning different layers (first, last, mix, non-existent, etc). Code is much more robust thanks to your feedback.

🤞