v2: oneof schema, typed builders, GraphFrames-style stub delivery

Implements Stage 1 of the protobuf architecture review (see deequ repo
docs/adr/0001..0005). The wire format moves from a string-discriminator
ConstraintMessage/AnalyzerMessage to typed `oneof` arms — one per Check
or Analyzer builder method — with reused `…Spec` payload submessages.

Schema-driven builder rewrites:
- pydeequ/v2/predicates.py — Predicate uses CompareOp enum; field
  renamed `operator` → `op`.
- pydeequ/v2/checks.py — each Check builder method (isComplete,
  hasPattern, isContainedIn, isLessThan, …) populates the matching
  oneof arm directly. The old _add_constraint(constraint_type, …)
  shape-shifter is gone.
- pydeequ/v2/analyzers.py — same pattern. Compliance gets its own
  ComplianceAnalyzerSpec (instance/predicate, no longer mislabeled as
  column/pattern). Pair-column constraints use named column_a/column_b.
- pydeequ/v2/suggestions.py — Rules enum values map to ConstraintRuleSet
  proto enum values directly. testset_seed=0 is now a legal user choice
  (proto3 `optional`).
- pydeequ/v2/profiles.py — `optional` low_cardinality_histogram_threshold.

Stub delivery (graphframes pattern, per ADR-0005):
- pydeequ/v2/proto/deequ_connect_pb2.py and _pb2.pyi are checked in.
  Refresh by running `DEEQU_PROTO_PATH=… python scripts/regen_proto.py`
  whenever the deequ schema changes; intermediate .proto is gitignored.
- scripts/regen_proto.py is a developer convenience, not a build hook.
- pyproject.toml gains grpcio-tools as a dev dependency for the regen
  script.
- .github/workflows/base.yml gains a (currently disabled) drift-check
  step that will activate once the matching deequ JAR ships .proto under
  META-INF/protobuf/.

Cross-cutting:
- WIRE_FORMAT_VERSION constant + field dropped from runners and proto
  __init__.py per ADR-0005 (Spark Connect's RelationPlugin contract uses
  the protobuf type URL as the version discriminator).
- tests/v2/test_unit.py rewritten to assert msg.WhichOneof("body")
  shape; 11 unit tests all pass.

Verified via end-to-end smoke test (Spark 3.5 Connect server + freshly
built deequ JAR + tutorials/data_quality_example_v2.py): 9 analyzers, 12
constraints (1 expected failure on duplicate detection), 10-column
profile, 19 constraint suggestions all produce correct DataFrames.

Author: chenliu0831

.github/workflows/base.yml

@@ -35,6 +35,10 @@ jobs:
           echo "SPARK_HOME=$PWD/spark-3.5.0-bin-hadoop3" >> $GITHUB_ENV
 
       - name: Download Deequ JAR
+        # The pinned JAR must match the schema in pydeequ/v2/proto/.
+        # When the schema changes, both the deequ JAR (built from the
+        # corresponding deequ branch) and this URL need to update in the
+        # same PR pair (per ADR-0004 in the deequ repo).
         run: |
           curl -L -o deequ_2.12-2.1.0b-spark-3.5.jar \
             https://github.com/awslabs/python-deequ/releases/download/v2.0.0b1/deequ_2.12-2.1.0b-spark-3.5.jar
@@ -46,6 +50,17 @@ jobs:
           poetry install
           poetry add "pyspark[connect]==3.5.0"
 
+      - name: Verify checked-in proto stubs are not stale
+        # ADR-0005 drift safeguard: regenerate the stubs from the JAR's
+        # bundled .proto and assert the diff is empty.
+        # NOTE: this requires a JAR that ships META-INF/protobuf/deequ_connect.proto
+        # — only deequ JARs built from the protobuf-stage1 branch onward do.
+        # Until the matching JAR is released, this step is skipped.
+        if: false
+        run: |
+          DEEQU_JAR_PATH=$PWD/deequ_2.12-2.1.0b-spark-3.5.jar poetry run python scripts/regen_proto.py
+          git diff --exit-code pydeequ/v2/proto/
+
       - name: Run V2 unit tests
         run: |
           poetry run pytest tests/v2/test_unit.py -v

.gitignore

@@ -148,5 +148,12 @@ dmypy.json
 # Cython debug symbols
 cython_debug/
 
-# DS_STORE 
+# Note: pydeequ/v2/proto/deequ_connect_pb2.py and _pb2.pyi are
+# CHECKED IN per ADR-0005 (graphframes pattern). Run scripts/regen_proto.py
+# to refresh them when the deequ schema changes.
+# The intermediate .proto extracted by regen_proto.py is NOT checked in —
+# the canonical schema lives in the deequ repo.
+pydeequ/v2/proto/deequ_connect.proto
+
+# DS_STORE
 .DS_Store