Implement safe MLX KV reuse with scoped GPU cache eviction

[mattt]

Alternative to #139

[Copilot]

Pull request overview

This PR extends MLXLanguageModel with session-scoped KV cache reuse and introduces a scoped GPU buffer-cache limit/eviction mechanism to improve multi-turn stability and reduce redundant prefill work in MLX-backed sessions.

Changes:

• Add MLX-specific GenerationOptions customizations (KV cache sizing + quantization knobs) and map them into MLX generate parameters.
• Implement per-LanguageModelSession KV cache storage/reuse plus a global “one active generation per session” gate and GPU cache-limit scoping.
• Add tests for multi-turn behavior and cache-clearing safety; document new MLX configuration in the README.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.

File	Description
Sources/AnyLanguageModel/Models/MLXLanguageModel.swift	Adds MLX custom options, session KV reuse, concurrency gate, and GPU cache-limit scoping/eviction.
Tests/AnyLanguageModelTests/MLXLanguageModelTests.swift	Adds multi-turn same-session test and cache-clear-then-respond test.
Tests/AnyLanguageModelTests/CustomGenerationOptionsTests.swift	Adds tests for MLXLanguageModel.CustomGenerationOptions initialization, integration, and Codable.
README.md	Documents MLX KV-cache tuning via GenerationOptions and GPU cache configuration via MLXLanguageModel init.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 6 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.