feat: [Orchestration] Support Fallback Modules

[Jonas-Isr]

Context

AI/ai-sdk-java-backlog#342

This PR introduces support for using fallback modules when using orchestration chat completion. Instead of a single OrchestrationModuleConfig object, users can now add multiple of these module configs to a chat completion call (streaming or non-streaming). The configs are tried in the order they are given and if a call with a config failed, the next in line is used as fallback.

This PR also introduces a convenience layer to do this:

var prompt = new OrchestrationPrompt("your prompt");

var workingConfig = new OrchestrationModuleConfig().withLlmConfig(GPT_4O_MINI);
var secondWorkingConfig = new OrchestrationModuleConfig().withLlmConfig(GPT_4O);
var brokenConfig =
    new OrchestrationModuleConfig()
        .withLlmConfig(new OrchestrationAiModel("broken_name", Map.of(), "latest"));

var fallbackConfigs = new OrchestrationModuleConfig[] {workingConfig, secondWorkingConfig};

OrchestrationChatResponse result = client.chatCompletion(prompt, brokenConfig, fallbackConfigs);
// or:
// OrchestrationChatResponse result = client.chatCompletion(prompt, brokenConfig, workingConfig, secondWorkingConfig);

In the above example, the call with brokenConfig will return an error so another call with workingConfig is performed and the (positive) result of this call is returned.

Feature scope:

• test low-level code
• add convenience layer
• unit and e2e tests

Definition of Done

• Functionality scope stated & covered
• Tests cover the scope above
• Error handling created / updated & covered by the tests above
• Aligned changes with the JavaScript SDK
• Documentation updated
  • Java: Orchestration Fallback Modules ai-sdk#377
• Release notes updated

[Jonas-Isr]

The stream options are taken from the first config only. (I don't think there is a way to change this.)

[CharlesDuboisSAP]

LGTM

[CharlesDuboisSAP]

Grounding in testFallbackModules?

[Jonas-Isr]

I added this here to be able to test that the additional configs (like grounding or input filter) are correctly propagated when using the fallback modules. Since this is not an e2e test, this will not add flakyness.