fix: remove dead Spec.layer_sql/stat_sql bindings (#15)

* fix: remove dead Spec.layer_sql/stat_sql bindings

These methods mirrored a vestigial Rust API. Since ggsql PR #82 (the
scale-syntax rewrite, first released in v0.1.6), Spec::new() has
populated the backing fields with `vec![None; ...]` — the methods have
always returned None in every released crate. ggsql PR #569 (the
dialect/SQL-construction centralization) removed them from the Rust API
entirely, so they no longer compile against ggsql main.

Nothing consumes them: the only usage in this repo was a smoke test
asserting the calls don't crash, and querychat (R and Python) never
calls them. Removing them restores forward-compatibility with ggsql
main at zero functional cost.

Verified by building and testing against both crates.io ggsql 0.5.2
and a local ggsql main checkout.

* docs: changelog entry for Spec.layer_sql/stat_sql removal

Author: cpsievert

CHANGELOG.md

@@ -7,6 +7,10 @@
 - Enabled the crate's `spatial` feature, so `DRAW spatial` map layers now render through the Vega-Lite writer instead of raising "Spatial visualization requires the 'spatial' feature to be enabled". Spatial layers require DuckDB's `spatial` extension at runtime: ggsql issues `LOAD spatial` automatically but never `INSTALL` it, so run `INSTALL spatial` once per machine (e.g. via `DuckDBReader.execute_sql()`; the download is cached under `~/.duckdb/extensions`). Works with WKB geometry columns and the built-in `ggsql:world` dataset, including named map projections such as `PROJECT TO equal_earth`.
 - New `CachingReader.execute_sql_cached()` method that runs SQL on the reader's private in-memory cache connection — the compute surface where all ggsql-generated derived SQL executes — instead of the primary data source. This is the surface for per-connection setup that derived SQL depends on: e.g. `INSTALL`/`LOAD` of DuckDB extensions such as `spatial`, which would fail on a locked-down primary and take no effect on the cache even where it succeeds.
 
+### Removed
+
+- Removed `Spec.layer_sql()` and `Spec.stat_sql()`. These mirrored a vestigial Rust API that returned `None` in every released crate since v0.1.6 (the backing fields were never populated after the scale-syntax rewrite) and was removed upstream in ggsql PR #569. No functionality is lost.
+
 ### Changed
 
 - Data now crosses the Python↔Rust boundary via the zero-copy Arrow C Data Interface instead of Arrow IPC serialization, reducing memory copies and peak RSS for large results (on a 1M-row × 50-col register+read microbenchmark: ~5.7× faster, ~1 GB lower peak RSS). Set `GGSQL_ARROW_IPC=1` to restore the previous IPC behavior. The minimum pyarrow version is now 15.

src/lib.rs

@@ -1255,36 +1255,6 @@ impl PySpec {
             .transpose()
     }
 
-    /// Layer filter/source query, or None if using global data.
-    ///
-    /// Parameters
-    /// ----------
-    /// index : int
-    ///     The layer index (0-based).
-    ///
-    /// Returns
-    /// -------
-    /// str | None
-    ///     The filter SQL query, or None if the layer uses global data directly.
-    fn layer_sql(&self, index: usize) -> Option<String> {
-        self.inner.layer_sql(index).map(|s| s.to_string())
-    }
-
-    /// Stat transform query, or None if no stat transform.
-    ///
-    /// Parameters
-    /// ----------
-    /// index : int
-    ///     The layer index (0-based).
-    ///
-    /// Returns
-    /// -------
-    /// str | None
-    ///     The stat transform SQL query, or None if no stat transform.
-    fn stat_sql(&self, index: usize) -> Option<String> {
-        self.inner.stat_sql(index).map(|s| s.to_string())
-    }
-
     /// Validation warnings from preparation.
     ///
     /// Returns

tests/test_ggsql.py

@@ -543,8 +543,6 @@ def test_can_introspect_spec(self):
         # Layer-specific accessors (may return None)
         _ = spec.layer_data(0)
         _ = spec.stat_data(0)
-        _ = spec.layer_sql(0)
-        _ = spec.stat_sql(0)
 
 
 class TestCustomReader: