refactor: move model pattern defaults to get-ai-gateway.sh

[johnlanni]

Motivation

Currently, model pattern defaults (e.g., MOONSHOT_MODELS="moonshot-*,kimi-*") are hardcoded in ai-gateway.sh which runs inside the container. When new models are released, updating these defaults requires rebuilding and redistributing the container image.

Changes

1. Move defaults to get-ai-gateway.sh

• Add all model pattern defaults in the resetEnv() function
• Defaults are now in the host script, making them easy to update via git

2. Update writeConfiguration()

• Persist all *_MODELS variables to the config file
• Ensure defaults are passed to the container via --env-file

3. Simplify ai-gateway.sh

• Remove hardcoded defaults (e.g., ${MOONSHOT_MODELS:-moonshot-*,kimi-*})
• Use environment variables directly (e.g., ${MOONSHOT_MODELS})
• Container now relies on externally-configured values

Benefits

✅ No container rebuild needed - Update model patterns by editing get-ai-gateway.sh
✅ Single source of truth - Defaults managed in one place (get-ai-gateway.sh)
✅ Version controlled - Model updates tracked in git history
✅ Backward compatible - Existing installations continue to work
✅ User override supported - Users can still pass --xxx-models flags

Example

# Defaults now in get-ai-gateway.sh (line ~580)
: "${MOONSHOT_MODELS:=moonshot-*,kimi-*}"

# When Moonshot releases kimi-k2, just update the script:
: "${MOONSHOT_MODELS:=moonshot-*,kimi-*,kimi-k2}"

# No need to rebuild containers!

Testing

• Verified existing installations continue to work
• Tested with default model patterns
• Tested with user-overridden patterns via CLI flags
• Confirmed config file persistence

Related Issue

Addresses the need to update model patterns frequently without image rebuilds (e.g., when providers release new models like Kimi-K1).