Fix MinicpmV model converter and clip to avoid using hardcode.

[gryffindor-rr]

Make sure to read the contributing guidelines before submitting a PR
Tested by llama-mtmd-cli with multiple minicpmv models.

[ngxson]

Please create PR from a non-master branch next time, otherwise I cannot push my corrections directly to this PR

[gryffindor-rr]

Please create PR from a non-master branch next time, otherwise I cannot push my corrections directly to this PR

sure. what branch will you suggest to use? shall I create a new one say 'minicpmv'?

[CISC]

This needs to be updated for MiniCPM-V 4.0 that was just merged.

[gryffindor-rr]

This needs to be updated for MiniCPM-V 4.0 that was just merged.

done.
convert models by following instructions listed in https://github.com/ggml-org/llama.cpp/blob/master/docs/multimodal/minicpmv4.0.md

example output:
./build/bin/llama-mtmd-cli -m ~/models/MiniCPM-V-4/model/Model-3.6B-F16.gguf --mmproj ~/models/MiniCPM-V-4/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image ~/codeup/edge_infer_sdk/tests/data/cat.png -p 'Describe image' -ngl 99
main: loading model: /cache/zhanglei/models/MiniCPM-V-4/model/Model-3.6B-F16.gguf
encoding image slice...
image slice encoded in 99 ms
decoding image batch 1/1, n_tokens_batch = 64
image decoded (batch 1/1) in 8 ms

The image features a close-up of a cat. The feline has distinct tabby markings on its fur, characterized by the distinctive stripes and swirls typical of the tabby pattern found in many domestic cats.
The background is blurred but appears to be an indoor setting with neutral colors that do not detract from the subject.
There are no visible texts or additional objects within this frame. The overall composition focuses solely on capturing the detailed features and calm demeanor of the cat as it gazes slightly upwards.

llama_perf_context_print: load time = 1823.19 ms
llama_perf_context_print: prompt eval time = 429.45 ms / 77 tokens ( 5.58 ms per token, 179.30 tokens per second)
llama_perf_context_print: eval time = 960.62 ms / 110 runs ( 8.73 ms per token, 114.51 tokens per second)
llama_perf_context_print: total time = 1852.51 ms / 187 tokens
llama_perf_context_print: graphs reused = 106

[CISC]

Can merge when @ngxson approves.

[CISC]

@ngxson gentle ping