FromRow impls are untestable because tokio_postgres::Row cannot be constructed

[0w3n-d]

Status

FromRow in crates/database/src/postgres/postgres_db_row_parsing.rs maps
query results to domain types across roughly 98 row.get / row.try_get
calls. None of it is unit tested, and it cannot be: every impl takes
&tokio_postgres::Row, and Row has no public constructor, so a test cannot
fabricate one. The only way to reach this code today is a live query, which
puts the workspace's most fiddly parsing logic behind an external service (see
#545).

Target structure: a narrow accessor trait so the parsing logic is independent
of where the values come from.

trait RowSource {
    fn get<'a, T: FromSql<'a>>(&'a self, col: &str) -> Result<T, DatabaseError>;
}

Implemented for &tokio_postgres::Row in production, and for a
HashMap-backed fake in tests. Each FromRow impl then becomes a
dependency-free unit test: build a fake row, parse it, assert the domain type,
and assert the error when a column is missing or has the wrong type.

Why now

Column-name typos, wrong types and missing null handling are exactly the bugs
this code has, and today none of them can be caught without a database. It
also blocks making the parsing logic reusable outside helix-database.

Behavior must not change

The production path stays &Row, only reached through the new trait. The
integration tests in postgres_db_service_tests.rs cover the round trip and
must keep passing unchanged — they are the characterisation tests for this
refactor, so #545 Step 1 should land first so they can be run deliberately.

Affected surface

crates/database/src/postgres/postgres_db_row_parsing.rs, and the call sites
that construct domain types from rows.

Steps (each becomes one PR)

• Step 1: Add the RowSource trait, implement it for &tokio_postgres::Row, and add the test fake. Migrate one FromRow impl as the worked example, with its unit tests. (tests: unit tests for the migrated impl, including the missing-column and wrong-type cases) (PR: )
• Step 2: Migrate the remaining FromRow impls, a few per PR, each with unit tests. (tests: per impl) (PR: )

Open questions

Should RowSource be indexed by column name only, or by position too? Name
only is safer and matches how the current impls read, but a couple of sites
may use positional access.