feat(vertexai): support meta's llama models

[CorieW]

Support Meta's Llama models in VertexAI.

Testing

Todo: Evaluating options for testing this, as requires deploying infrastructure.

[gemini-code-assist]

Code Review

This pull request introduces support for Meta (Llama) models within the Vertex AI plugin by implementing an OpenAI-compatible API client and the necessary request/response mapping logic. Key changes include the addition of the VertexAiMetaOptions class for model configuration, the meta method for model referencing, and updates to the plugin's discovery logic to include Meta models. Review feedback identifies critical syntax errors in map literals where an invalid ?variable syntax was used instead of collection if statements, as well as a potential null-safety issue when joining text parts.

[gemini-code-assist]

The ?variable syntax used in this map literal is not valid Dart. To conditionally include entries in a map based on whether a value is null, you should use the collection if syntax.

Suggested change

	'version': ?version,
	'temperature': ?temperature,
	'topP': ?topP,
	'maxTokens': ?maxTokens,
	'stop': ?stop,
	'presencePenalty': ?presencePenalty,
	'frequencyPenalty': ?frequencyPenalty,
	'seed': ?seed,
	'user': ?user,
	'llamaGuard': ?llamaGuard,
	if (version != null) 'version': version,
	if (temperature != null) 'temperature': temperature,
	if (topP != null) 'topP': topP,
	if (maxTokens != null) 'maxTokens': maxTokens,
	if (stop != null) 'stop': stop,
	if (presencePenalty != null) 'presencePenalty': presencePenalty,
	if (frequencyPenalty != null) 'frequencyPenalty': frequencyPenalty,
	if (seed != null) 'seed': seed,
	if (user != null) 'user': user,
	if (llamaGuard != null) 'llamaGuard': llamaGuard,

[gemini-code-assist]

The ?variable syntax used in this map literal is not valid Dart. To conditionally include entries in a map based on whether a value is null, you should use the collection if syntax.

Suggested change

	'temperature': ?options.temperature,
	'top_p': ?options.topP,
	'max_tokens': ?options.maxTokens,
	'stop': ?options.stop,
	'presence_penalty': ?options.presencePenalty,
	'frequency_penalty': ?options.frequencyPenalty,
	'seed': ?options.seed,
	'user': ?options.user,
	if (options.temperature != null) 'temperature': options.temperature,
	if (options.topP != null) 'top_p': options.topP,
	if (options.maxTokens != null) 'max_tokens': options.maxTokens,
	if (options.stop != null) 'stop': options.stop,
	if (options.presencePenalty != null) 'presence_penalty': options.presencePenalty,
	if (options.frequencyPenalty != null) 'frequency_penalty': options.frequencyPenalty,
	if (options.seed != null) 'seed': options.seed,
	if (options.user != null) 'user': options.user,

[gemini-code-assist]

While parts.every((part) => part.isText) ensures all parts are text parts, part.text on the base Part class typically returns a nullable String?. Using join on an iterable of nullable strings can result in the literal string "null" being included in the output if any value is null. It is safer to cast the parts to TextPart to ensure non-nullability.

Suggested change

	return parts.map((part) => part.text).join('\n');
	return parts.cast<TextPart>().map((part) => part.text).join('\n');