feat(M6a): Pipeline + Profile API (additive) for v1.0

CHANGELOG.md

@@ -75,6 +75,26 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Miri-friendly test target `tests/miri_resegmentation.rs` covering
   no-overlap, single-overlap, and centroid math paths.
 
+### Added (M6a — Pipeline + Profile API, additive)
+- `polyvoice::pipeline_v2` module: `Pipeline::builder()` returning
+  `PipelineBuilder` with `.profile(Profile::Mobile|Balanced|Custom)`,
+  `.with_models_from(ModelRegistry)`, `.with_segmenter/embedder/clusterer/resegmenter()`,
+  and a validated `.build()`. `PipelineConfig`, `ClustererKind`,
+  `ExecutionProvider`, and `ConfigError` all per spec §5.2/§5.4.
+- `Pipeline::run(&samples, SampleRate)` orchestrates M1 segmenter → M2
+  embedder (with `apply_overlap_mask` per primary chunk) → M3 clusterer →
+  M4 resegmenter → legacy `merge_segments` → `DiarizationResult`.
+- New Cargo feature `pipeline_v2` (in default features). Requires
+  `onnx + segmentation + embedder + clusterer + resegmentation`; missing
+  any of these triggers a `compile_error!` with an actionable message.
+- Public re-exports `polyvoice::PipelineV2`, `PipelineBuilder`,
+  `PipelineConfig`, `ClustererKind`, `ExecutionProvider`, `ConfigError`,
+  `PipelineV2Error`. Legacy `polyvoice::Pipeline` is unchanged; M6b will
+  rename `pipeline_v2 → pipeline` and remove the legacy code path.
+- Synthetic integration test on Custom profile (`tests/pipeline_v2_synthetic_test.rs`,
+  7 tests) + `#[ignore]` E2E test on Balanced profile via `ModelRegistry`
+  (`tests/pipeline_v2_e2e_test.rs`).
+
 ### Added (M5 — INT8 quantization)
 - New scripts: `download-voxconverse-dev.sh`, `download-voxceleb1-subset.sh`,
   `quantize_models.py` + `quantize-models.sh`, `validate_int8.py` +

Cargo.toml

@@ -12,7 +12,7 @@ categories = ["multimedia::audio", "science"]
 crate-type = ["cdylib", "rlib"]
 
 [features]
-default = ["spectral", "segmentation", "embedder", "clusterer", "resegmentation"]
+default = ["spectral", "segmentation", "embedder", "clusterer", "resegmentation", "pipeline_v2"]
 # ONNX-based embedding extractors (pulls ort).
 onnx = ["dep:ort"]
 # C FFI bindings.
@@ -52,6 +52,13 @@ clusterer = []
 # caller (M6 Pipeline wires the embedder pool into this).
 resegmentation = []
 
+# v1.0 Pipeline + Profile builder API (additive in M6a; replaces legacy in M6b).
+# Requires the full M1–M5 stack — the builder loads ONNX through ModelRegistry
+# (download feature) for Mobile/Balanced profiles. The module gates itself
+# with a `compile_error!` if onnx/segmentation/embedder/clusterer/resegmentation
+# are missing, so half-wired feature combos cannot ship.
+pipeline_v2 = ["download"]
+
 # Model profile bundles. M0 only emits metadata; CLI uses these to pick which
 # manifest entries to ensure(). M5 will introduce INT8 model variants.
 profile-mobile = []