chore(doc): clarify compile-time verification and case conversion behavior

sqlx-core/src/from_row.rs

@@ -65,6 +65,14 @@ use crate::{error::Error, row::Row};
 /// reason), `lowercase`, `UPPERCASE`, `camelCase`, `PascalCase`, `SCREAMING_SNAKE_CASE` and `kebab-case`.
 /// The styling of each option is intended to be an example of its behavior.
 ///
+/// Case conversion is handled by the `heck` crate.
+/// See [its documentation](https://docs.rs/heck/0.5.0/heck/#definition-of-a-word-boundary)
+/// for details.
+///
+/// Note that numbers are *not* considered separate words.
+/// For example, `Foo1` to snake case would be `foo1`, *not* `foo_1`.
+/// See [this issue](https://github.com/launchbadge/sqlx/issues/3864) for discussion.
+///
 /// #### `default`
 ///
 /// When your struct contains a field that is not present in your query,

sqlx-core/src/types/mod.rs

@@ -99,8 +99,11 @@ pub use bstr::{BStr, BString};
 ///
 /// ## Compile-time verification
 ///
-/// With compile-time verification, the use of type overrides is currently required to make
-/// use of any user-defined types.
+/// Type definitions are *not* verified against the database at compile-time.
+/// The [`query!()`](macro.query.html) macros have no implicit knowledge of user-defined types.
+///
+/// When using custom types in query parameters or output columns with `query!()`,
+/// the use of [type overrides](macro.query.html#type-overrides-bind-parameters-postgres-only) is required.
 ///
 /// ```rust,ignore
 /// struct MyUser { id: UserId, name: String }

src/macros/mod.rs

@@ -249,7 +249,7 @@
 ///
 /// ##### Force a Different/Custom Type
 /// Selecting a column `foo as "foo: T"` (Postgres / SQLite) or `` foo as `foo: T` `` (MySQL)
-/// overrides the inferred type which is useful when selecting user-defined custom types
+/// overrides the inferred type which is useful when selecting user-defined [custom types][crate::Type#compile-time-verification]
 /// (dynamic type checking is still done so if the types are incompatible this will be an error
 /// at runtime instead of compile-time). Note that this syntax alone doesn't override inferred nullability,
 /// but it is compatible with the forced not-null and forced nullable annotations: