Port ProviderConfig model/token overrides and client startup cleanup fix from reference implementation

[Copilot]

Resolves #398

---

Before the change?

• ProviderConfig had no way to override the model name sent to the provider API or control token limits
• CopilotClient leaked the CLI process if startup failed after process launch (e.g., protocol version mismatch)
• BYOK docs contained verbose identity limitation details removed upstream

After the change?

• ProviderConfig gains modelId, wireModel, maxInputTokens, maxOutputTokens fields:

new ProviderConfig()
    .setType("openai")
    .setBaseUrl("https://api.openai.com/v1")
    .setApiKey("sk-...")
    .setModelId("gpt-4o")              // runtime config lookup
    .setWireModel("my-finetune-v3")    // actual name sent to provider
    .setMaxInputTokens(100_000)        // serialized as "maxPromptTokens"
    .setMaxOutputTokens(4096)

• CopilotClient.startCoreBody() now destroys the CLI process on partial startup failure, preventing resource leaks
• BYOK identity limitations section simplified to match upstream

Reference implementation commits ported: 58cf64d, 3c622b9, 100390a

Not ported (intentionally): 1832d0a, 79dc615, 08b486d, c9433c4, c9c3648, ad0a9aa — test infrastructure, .NET-specific, or generated code (version synced via finish script).

Pull request checklist

• Tests for the changes have been added (for bug fixes / features)
• Docs have been reviewed and added / updated if needed (for bug fixes / features)
• mvn spotless:apply has been run to format the code
• mvn clean verify passes locally

Does this introduce a breaking change?

• Yes
• No

---