[codex] Switch default model to Gemma 4 and add CI

[amariichi]

Summary

• switch the default local model configuration from Qwen3.5 to Gemma 4 E4B with mmproj-F16
• set single-user llama-server default parallelism to 1 and document the new defaults
• add a minimal GitHub Actions CI workflow for Python and shell syntax checks
• bump the app version to 6.0.0 to reflect the breaking default-model change

Validation

• python3 -m compileall app
• bash -n start.sh
• manual local smoke check with Gemma 4 GGUF + mmproj