Fix: always send stream flag explicitly in LLM request body

[Kayaba-Attribution]

Description

Fixes a bug where Ollama and other OpenAI-compatible providers would silently fail AI enhancement when streaming was disabled.

When `config.streaming` was `false`, the `stream` key was omitted entirely from the JSON request body. Providers like Ollama treat an absent `stream` key as `true` — so they returned SSE-encoded chunks, the non-streaming parser couldn't handle them, and the call failed with `LLMError.invalidResponse`. The enhancement would silently fall back to the raw transcript.

The fix is to always send the key explicitly in both `buildChatCompletionsBody` and `buildResponsesBody`:

```swift
// Before — key absent when false; Ollama reads absence as true
if config.streaming {
body["stream"] = true
}

// After — always explicit
body["stream"] = config.streaming
```

Type of Change

• 🐞 Bug fix (non-breaking change which fixes an issue)

Related Issues

• Closes [🐞 BUG] AI Enhancements fail when provider defaults to streaming (stream flag omitted from request body) #295

Testing

• Tested on Apple Silicon Mac (M-series)
• Tested on macOS Sequoia 15
• 4 unit tests added in `LLMClientRequestBodyTests` asserting the `stream` key is present and correct for both endpoints (`/chat/completions` and `/responses`) and both streaming states
• Tested on Intel Mac
• Ran linter locally: `swiftlint --strict --config .swiftlint.yml Sources` — 0 violations
• Ran formatter locally: `swiftformat --config .swiftformat Sources/Fluid/Services/LLMClient.swift` — 1 cosmetic fix (`//` → `///` on an existing comment)

Notes

• `buildChatCompletionsBody` and `buildResponsesBody` were changed from `private` to `internal` to allow direct unit testing. No behaviour changes outside the module.
• The `stream: false` fix is most impactful for Ollama and self-hosted providers. OpenAI and Groq already default to non-streaming when the key is absent, so this is a no-op for them.