Shortnames are not compatible with sandbox

[BansheeHero]

Issue Description

• Severity: Low, the models can be selected later in the agents.
• Issue: The paths between serve and sandbox do not match when using shortnames (even in dryrun) .
• Full name: ibm-granite/granite-3.3-8b-instruct-GGUF
• Shortname: granite
• Name provided to Llama.cpp: destination=/mnt/models/huggingface-ibm-granite-granite-3.3-8b-instruct-GGUF-latest.gguf

Steps to reproduce the issue

1. Serve your stored models: ramalama --dryrun serve | grep granite
2. Run any sandbox against the served models using a shortname: ramalama --dryrun sandbox pi --url http://169.254.1.2:8080 granite | grep granite
3. Agent starts with a connection error.

Describe the results you received

Errors that model is not available.

podman run
    --rm
    --label ai.ramalama.engine=podman
    --label ai.ramalama.runtime=llama.cpp
    --label ai.ramalama.port=8080
    --label ai.ramalama.command=serve
    --mount=type=bind,src=/home/banshee/.local/share/ramalama/store/huggingface/ibm-granite/granite-3.3-8b-instruct-GGUF/blobs/sha256-77bcee066a76dcdd10d0d123c87e32c8ec2c74e31b6ffd87ebee49c9ac215dca,destination=/mnt/models/huggingface-ibm-granite-granite-3.3-8b-instruct-GGUF-latest.gguf,ro
    --mount=type=bind,src=/home/banshee/.local/share/ramalama/store/huggingface/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF/blobs/sha256-fadc3e5f8d42bf7e894a785b05082e47daee4df26680389817e2093056f088ad,destination=/mnt/models/huggingface-unsloth-Qwen3-Coder-30B-A3B-Instruct-GGUF-latest.gguf,ro
    --mount=type=bind,src=/home/banshee/.local/share/ramalama/store/huggingface/Qwen/Qwen2.5-Coder-7B-Instruct-GGUF/blobs/sha256-6b99ee26f4b1f887b25dbb45491ec158391ba7ba73dbc4c75ca9560d3da4493a,destination=/mnt/models/huggingface-Qwen-Qwen2.5-Coder-7B-Instruct-GGUF-Q6_K.gguf,ro
    --mount=type=bind,src=/home/banshee/.local/share/ramalama/store/huggingface/bartowski/Qwen_Qwen3.5-4B-GGUF/blobs/sha256-13c16f426047e2de38cd075bdade4a7bcbc8c774384876f677740cda65f8a983,destination=/mnt/models/huggingface-bartowski-Qwen_Qwen3.5-4B-GGUF-latest.gguf,ro
    --mount=type=bind,src=/home/banshee/.local/share/ramalama/store/huggingface/bartowski/Qwen_Qwen3.6-27B-GGUF/blobs/sha256-8739a0cbb80036e5dbdced2085f142b8ba86e3235db8b8039b3769fe5fc70843,destination=/mnt/models/huggingface-bartowski-Qwen_Qwen3.6-27B-GGUF-latest.gguf,ro quay.io/ramalama/ramalama:0.24 llama-server

podman run --rm -i 
    --label ai.ramalama.model=hf://ibm-granite/granite-3.3-8b-instruct-GGUF 
    --label ai.ramalama.engine=podman --label ai.ramalama.runtime=llama.cpp  
    --label ai.ramalama.port=8080 --label ai.ramalama.command=sandbox  
    --network container:None  
    --security-opt=label=disable --cap-drop=all --security-opt=no-new-privileges 
    --pull newer -t --name pi-None --env LLAMA_SERVER_URL=http://localhost:8080 quay.io/ramalama/pi-agent:0.24  
    --provider llama-server  
    --model ibm-granite/granite-3.3-8b-instruct-GGUF 

Describe the results you expected

Paths match between serve and sandbox

ramalama info output

#2929

ramalama list:

SHORTNAME       NAME                                                  MODIFIED   SIZE     
                hf://unsloth/gemma-4-E4B-it-GGUF                      3 days ago 5.56 GB  
                hf://unsloth/gemma-4-E2B-it-GGUF                      3 days ago 3.81 GB  
                hf://unsloth/gemma-4-12B-it-GGUF                      3 days ago 6.8 GB   
                hf://unsloth/Qwen3.6-35B-A3B-MTP-GGUF                 3 days ago 21.94 GB 
                hf://unsloth/Qwen3.5-9B-GGUF                          3 days ago 6.15 GB  
                hf://unsloth/Qwen3.5-4B-GGUF                          3 days ago 3.18 GB  
                hf://unsloth/Qwen3.5-2B-GGUF                          3 days ago 1.82 GB  
                hf://unsloth/Qwen3.5-0.8B-GGUF                        3 days ago 703.35 MB
qwen3-coder:30b hf://unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF        2 days ago 17.28 GB 

Additional information

I think shortnames should be easier to edit and maintained - Especially since a lot of them follow a formula.
This might be an issue with upstream, apologies if I made a mistake. I do believe this is part of the features ramalama offers

[BansheeHero]

Suggested test to catch this:

• I didn't want to make a PR just for a test - I hear that is something to do when reporting issues upstream.
• Current test only checks the shortname: "gpt-oss" without quant or latest
• There are already pre-defined shortnames that follow this scenario: granite
• Output could be tested against the full argument: granite -> --model ibm-granite/granite-3.3-8b-instruct-GGUF-latest

Or (my preferred solution) do not add the latest to serve when quant is missing:
--mount=type=bind,src=/home/banshee/.local/share/ramalama/store/huggingface/ibm-granite/granite-3.3-8b-instruct-GGUF/blobs/sha256-77bcee066a76dcdd10d0d123c87e32c8ec2c74e31b6ffd87ebee49c9ac215dca,destination=/mnt/models/huggingface-ibm-granite-granite-3.3-8b-instruct-GGUF-latest.gguf