refactor(runtime)!: split model runtime protocols by capability

[laipz8200]

Important

1. Make sure you have read our contribution guidelines
2. Search existing issues and pull requests to confirm this change is not a duplicate
3. Open or identify the issue this pull request resolves or advances
4. Use a Conventional Commits title for this pull request, and mark breaking changes with !
5. Remember that the pull request title will become the squash merge commit message
6. If CLA Assistant prompts you, sign CLA.md in the pull request conversation

Related Issue

Closes #56

Summary

This refactor breaks the monolithic model runtime contract into capability-specific protocols and moves those protocol definitions under graphon.model_runtime.protocols.

The change narrows each model wrapper to the runtime surface it actually consumes, updates Slim runtime integrations to use the new protocol package, exports the new protocol set from graphon.protocols, and makes ModelProviderFactory a provider-only abstraction backed by ModelProviderRuntime.

Breaking Changes

• The top-level module path graphon.model_runtime.runtime is removed in favor of graphon.model_runtime.protocols.runtime.
• Capability protocol module paths move from graphon.model_runtime.<model>_runtime to graphon.model_runtime.protocols.<model>_runtime.
• ModelProviderFactory now accepts runtime= backed by ModelProviderRuntime instead of model_runtime= backed by the aggregate ModelRuntime.
• ModelProviderFactory.get_model_type_instance() is removed. Callers that need invocation behavior must instantiate the appropriate model wrapper directly against the matching capability-specific runtime.
• Custom runtime implementations that previously targeted the old monolithic module paths must update their imports to the new protocol package.
• Downstream type integrations should now target ModelProviderRuntime, LLMModelRuntime, TextEmbeddingModelRuntime, RerankModelRuntime, SpeechToTextModelRuntime, ModerationModelRuntime, or TTSModelRuntime directly where appropriate instead of assuming a single required interface.
• Typed integrations around SlimPreparedLLM and other LLM-facing helpers now depend on the LLM-specific runtime protocol package path.

Downstream Checklist

• Replace imports from graphon.model_runtime.runtime with graphon.model_runtime.protocols.runtime where a full aggregate runtime is still required.
• Replace imports from graphon.model_runtime.llm_runtime, text_embedding_runtime, rerank_runtime, speech_to_text_runtime, moderation_runtime, tts_runtime, and provider_runtime with their graphon.model_runtime.protocols.* equivalents.
• Update ModelProviderFactory(...) call sites that passed model_runtime= to use runtime=.
• Replace ModelProviderFactory.get_model_type_instance() usage with direct construction of LargeLanguageModel, TextEmbeddingModel, RerankModel, Speech2TextModel, ModerationModel, or TTSModel against the corresponding capability-specific runtime.
• Update custom runtimes to implement only the capability-specific protocols they actually support.
• Remove placeholder or stub methods that existed only to satisfy unrelated parts of the previous monolithic interface.
• Update any factories, dependency injection bindings, or adapter registries that referenced the removed top-level protocol module paths.
• Review typed call sites around SlimPreparedLLM, model wrappers, and provider factories to ensure they point at the new protocol package and provider-only factory boundary.
• Search internal docs, examples, and downstream libraries for the removed module paths and migrate them to the new package structure.

Checklist

• This pull request links the issue it resolves or advances
• This pull request title follows Conventional Commits, and any breaking change is marked with !
• If CLA Assistant prompted me, I signed CLA.md in the pull request conversation