fix(writer): align ducklake_column / ducklake_data_file schema with the DuckLake spec (#116)

* fix(writer): align ducklake_column / ducklake_data_file with the spec

`SqliteMetadataWriter` was producing a `ducklake_column` table that
lacked `parent_column`, `initial_default`, `default_value`,
`default_value_type`, and `default_value_dialect`. The reader's
SQL_GET_TABLE_COLUMNS query (added in PR #89) projects
`parent_column`, so any catalog written by `SqliteMetadataWriter` was
unreadable through the metadata provider, panicking with
"no such column: parent_column".

The bug shipped because CI ran `cargo test` without `write-sqlite`
features, and both `tests/sql_write_tests.rs` and
`tests/sqlite_metadata_provider_test.rs` are gated behind features
that CI didn't enable — so the broken-catalog path compiled to a
zero-test binary in CI and was invisible.

Three changes:

1. `metadata_writer_sqlite.rs::SQL_CREATE_SCHEMA` adds the five
   missing `ducklake_column` columns, matching the upstream
   `ducklake_metadata_manager.cpp` schema definition. We don't
   populate the four `*_default*` columns today (we don't support
   nested types or column defaults yet), but the columns must exist
   so the reader query doesn't error.

2. `tests/sqlite_metadata_provider_test.rs::init_schema` is brought
   in line with the same spec. It also gains `row_id_start`,
   `record_count`, and `mapping_id` on `ducklake_data_file` — these
   are projected by SQL_GET_DATA_FILES since the row-lineage work
   landed in PR #115 and any sqlite-fixture test that exercises the
   data-file read path would have hit "no such column" without them.

3. `.github/workflows/ci.yml` enables `write-sqlite` on the test
   step so `sql_write_tests` and `sqlite_metadata_provider_test`
   actually run. Without this, the next analogous schema drift will
   ship invisibly again.

Verified by running `cargo test --features "skip-tests-with-docker
write-sqlite"` locally — 228 unit tests + every integration suite
green, including the 7 previously-broken `sql_write_tests` and 18
`sqlite_metadata_provider_test` cases.

Out of scope (separately broken with the same root cause, no CI
coverage today): the encryption, postgres, and mysql test fixtures
have analogous spec gaps. Worth a follow-up once CI for those
backends is set up.

* docs(ci): reword the write-sqlite comment so the intent isn't ambiguous

The earlier wording read like "enable write-sqlite so the file runs"
which could be parsed as either "we previously didn't run them and
now we are" (correct) or "we want to skip running them" (wrong).
Rephrased to make clear we're adding features so the tests become
visible to cargo, not the other way round.

* docs: trim the new comments to just the non-obvious WHY

Three comment blocks added earlier in this branch were verbose,
duplicated reasoning across files, or carried a PR reference that
will rot. Trimmed each to the smallest signal a future reader needs:

* CI: 6 lines → 3. Keep the "without this, cargo compiles to a
  zero-test binary" hazard; drop the cfg-syntax tutorial.

* metadata_writer_sqlite.rs: 8 lines → 5. Spell out who reads each
  group of columns (our reader vs DuckDB) without re-explaining the
  same thing twice.

* sqlite_metadata_provider_test.rs: 7 lines across two blocks → 1.
  Replace the duplicated "why columns exist" lecture with a single
  pointer to the writer file. Also drops the "since PR #115"
  reference per the don't-cite-PR-numbers convention — the linkage
  is now in the writer file's comment.

Author: anoop-narang

.github/workflows/ci.yml

@@ -83,11 +83,14 @@ jobs:
         run: cargo clippy --all-targets --all-features --no-deps -- -D warnings
 
       - name: Run tests
+        # `write-sqlite` pulls in the writer and sqlite-provider tests,
+        # which are `#![cfg]`-gated; without it cargo compiles them to
+        # zero-test binaries.
         run: |
           if [[ "$RUNNER_OS" == "macOS" ]]; then
-            cargo test --features skip-tests-with-docker
+            cargo test --features "skip-tests-with-docker write-sqlite"
           else
-            cargo test
+            cargo test --features "write-sqlite"
           fi
 
       - name: Check documentation

src/metadata_writer_sqlite.rs

@@ -50,6 +50,16 @@ CREATE TABLE IF NOT EXISTS ducklake_column (
     column_type VARCHAR NOT NULL,
     column_order INTEGER NOT NULL,
     nulls_allowed BOOLEAN DEFAULT 1,
+    -- Mirror the upstream DuckLake spec (ducklake_metadata_manager.cpp):
+    -- `parent_column` is projected by our reader's SQL_GET_TABLE_COLUMNS;
+    -- the four `*default*` columns are projected by DuckDB when it reads
+    -- catalogs we produce. We leave them NULL — no nested-type or
+    -- column-default writes yet.
+    initial_default VARCHAR,
+    default_value VARCHAR,
+    parent_column INTEGER,
+    default_value_type VARCHAR,
+    default_value_dialect VARCHAR,
     begin_snapshot INTEGER NOT NULL,
     end_snapshot INTEGER
 );

tests/sqlite_metadata_provider_test.rs

@@ -68,6 +68,7 @@ async fn init_schema(pool: &SqlitePool) -> anyhow::Result<()> {
     .execute(pool)
     .await?;
 
+    // Schema must match SQL_CREATE_SCHEMA in metadata_writer_sqlite.rs.
     sqlx::query(
         "CREATE TABLE IF NOT EXISTS ducklake_column (
             column_id INTEGER PRIMARY KEY,
@@ -76,6 +77,11 @@ async fn init_schema(pool: &SqlitePool) -> anyhow::Result<()> {
             column_type TEXT NOT NULL,
             column_order INTEGER NOT NULL,
             nulls_allowed INTEGER,
+            initial_default TEXT,
+            default_value TEXT,
+            parent_column INTEGER,
+            default_value_type TEXT,
+            default_value_dialect TEXT,
             begin_snapshot INTEGER NOT NULL DEFAULT 1,
             end_snapshot INTEGER,
             FOREIGN KEY (table_id) REFERENCES ducklake_table(table_id)
@@ -93,6 +99,9 @@ async fn init_schema(pool: &SqlitePool) -> anyhow::Result<()> {
             file_size_bytes INTEGER NOT NULL,
             footer_size INTEGER,
             encryption_key TEXT,
+            record_count INTEGER,
+            row_id_start INTEGER,
+            mapping_id INTEGER,
             begin_snapshot INTEGER NOT NULL DEFAULT 1,
             end_snapshot INTEGER,
             FOREIGN KEY (table_id) REFERENCES ducklake_table(table_id)