POC: iceberg-testing conformance fixtures (submodule)

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "iceberg-testing"]
+	path = iceberg-testing
+	url = https://github.com/sungwy/iceberg-testing.git

conformance/README.md

@@ -0,0 +1,26 @@
+# Conformance fixtures (POC — submodule wired, test deferred)
+
+This pins [`sungwy/iceberg-testing`](https://github.com/sungwy/iceberg-testing) as a
+git submodule at `iceberg-testing/`: language-neutral conformance fixtures for
+Apache Iceberg, modeled on `apache/parquet-testing`. Three surfaces:
+
+- `table-spec/types/` — type-string parse + exact re-serialize (canonical)
+- `table-spec/transforms/bucket/` — the Appendix B 32-bit bucket hash, as a
+  Known-Answer-Test, including byte-boundary decimals and a non-BMP string
+- `table-spec/delete-formats/` — positional and equality delete-file decode by
+  field-id
+
+A C++ (`ctest`) harness that walks these fixtures is **not included in this POC**.
+It was prepared in an environment without a C++ toolchain, so rather than commit an
+unverified, possibly non-compiling test, the wiring is left as a TODO. The PyIceberg
+fork has the worked example, including how a consumer keeps a local staged-adoption
+list for cases it does not yet satisfy:
+https://github.com/sungwy/iceberg-python/pull/1
+
+To consume: `git submodule update --init`, then walk `iceberg-testing/table-spec/**`
+and apply each surface README's assertion (parse + re-serialize; compute the bucket
+hash and compare; decode delete-file columns by field-id).
+
+The bucket Known-Answer-Test is the highest-signal surface to wire first: the
+expected hashes are taken from the spec, and the byte-boundary decimal cases
+(`-1.28`, `-327.68`) are where minimal two's-complement encoding has diverged.