feat: implement Clang-backed metadata adapter and registry for C++ inference enrichment

Author: apstenku123

cppmega_mlx/inference/__init__.py

@@ -41,13 +41,19 @@
     require_model_integrated_paged_attention,
 )
 from cppmega_mlx.inference.side_channels import (
+    AdapterUnavailableError,
     AdapterCapabilities,
+    ClangCodeMetadataAdapter,
     CodeMetadataAdapter,
     InferenceFailPolicy,
     InferenceSideChannelBuilder,
     InferenceSideChannelCacheComponents,
     InferenceSideChannelResult,
     TokenMetadata,
+    UnavailableCodeMetadataAdapter,
+    builtin_code_metadata_adapters,
+    get_builtin_code_metadata_adapter,
+    normalize_code_language,
 )
 from cppmega_mlx.inference.speculative_decode import (
     speculative_acceptance,
@@ -61,7 +67,9 @@
     "ContiguousKVCacheConfig",
     "ContinuousBatchScheduler",
     "GenerationChunk",
+    "AdapterUnavailableError",
     "AdapterCapabilities",
+    "ClangCodeMetadataAdapter",
     "CodeMetadataAdapter",
     "InferenceFailPolicy",
     "InferenceQuantizationConfig",
@@ -75,17 +83,21 @@
     "SchedulerOutput",
     "SequenceRequest",
     "TokenMetadata",
+    "UnavailableCodeMetadataAdapter",
     "build_prompt_cache",
     "build_fim_prompt_ids",
     "build_paged_block_table",
+    "builtin_code_metadata_adapters",
     "clone_contiguous_kv_cache",
     "generate_tokens",
     "generate_tokens_with_prompt_cache",
     "generate_tokens_with_kv_cache",
+    "get_builtin_code_metadata_adapter",
     "kv_cache_position",
     "make_contiguous_kv_cache",
     "make_quantized_kv_cache",
     "next_token_logits",
+    "normalize_code_language",
     "prefill_contiguous_kv_cache",
     "quantize_kv_cache",
     "quantize_linear_for_inference",