benches: package as kernbench.benches, add @bench registry + list subcommand

Move benches/ -> src/kernbench/benches/ and src/kernbench/cli/probe.py ->
src/kernbench/probes/probe.py. Each bench self-registers via
@bench(name=..., description=...); kernbench list enumerates benches
with auto-assigned indices, --bench accepts kebab-case name or numeric
index. Audit at package-import time fails if any non-underscore module
forgets the decorator. ADR-0010 (EN + KO) updated to reflect the new
resolver path, list subcommand, and probes package separation.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

tests/test_bench_registry.py

@@ -0,0 +1,95 @@
+"""Tests for kernbench.benches.registry — @bench decorator + resolve/list."""
+from __future__ import annotations
+
+import pytest
+
+from kernbench.benches import registry
+
+
+EXPECTED_NAMES = [
+    "ccl-allreduce",
+    "gemm-single-pe",
+    "gpt3-qkv",
+    "ipcq-allreduce",
+    "matmul-composite",
+    "qkv-gemm",
+    "qkv-gemm-multi-pe",
+    "va-offset-verify",
+]
+
+
+def test_registry_lists_all_benches():
+    specs = registry.list_all()
+    names = [s.name for s in specs]
+    assert names == EXPECTED_NAMES
+
+
+def test_registry_indices_are_1_based_sorted_by_name():
+    specs = registry.list_all()
+    assert [s.index for s in specs] == list(range(1, len(EXPECTED_NAMES) + 1))
+    assert sorted(s.name for s in specs) == [s.name for s in specs]
+
+
+def test_resolve_by_name_returns_spec():
+    spec = registry.resolve("gemm-single-pe")
+    assert spec.name == "gemm-single-pe"
+    assert callable(spec.run)
+    assert spec.description.strip()
+
+
+def test_resolve_by_index_string_matches_list_order():
+    specs = registry.list_all()
+    third = specs[2]
+    resolved = registry.resolve(str(third.index))
+    assert resolved is third
+
+
+def test_resolve_unknown_name_raises():
+    with pytest.raises(ValueError, match="kernbench list"):
+        registry.resolve("does-not-exist")
+
+
+def test_resolve_unknown_index_raises():
+    with pytest.raises(ValueError, match="kernbench list"):
+        registry.resolve("99")
+
+
+def test_resolve_empty_identifier_raises():
+    with pytest.raises(ValueError):
+        registry.resolve("")
+
+
+def test_bench_decorator_rejects_invalid_name():
+    with pytest.raises(ValueError, match="kebab-case"):
+        registry.bench(name="Invalid_Name", description="x")
+
+
+def test_bench_decorator_rejects_empty_description():
+    with pytest.raises(ValueError, match="non-empty"):
+        registry.bench(name="ok-name", description="   ")
+
+
+def test_audit_raises_on_missing_decorator():
+    with pytest.raises(RuntimeError, match="missing @bench decorator"):
+        registry._audit_modules(
+            imported=["kernbench.benches.fake_no_dec", "kernbench.benches.real"],
+            registered={"kernbench.benches.real"},
+        )
+
+
+def test_audit_passes_when_all_registered():
+    registry._audit_modules(
+        imported=["kernbench.benches.a", "kernbench.benches.b"],
+        registered={"kernbench.benches.a", "kernbench.benches.b"},
+    )
+
+
+def test_duplicate_name_at_finalize_fails(monkeypatch):
+    """_finalize() rejects two pending entries with the same name."""
+    monkeypatch.setattr(registry, "_PENDING", [
+        ("dup", "d1", lambda: None),
+        ("dup", "d2", lambda: None),
+    ])
+    monkeypatch.setattr(registry, "_REGISTRY", {})
+    with pytest.raises(RuntimeError, match="duplicate bench name"):
+        registry._finalize()