fix(linux-sqlite): let SQLite decide what holds a dropped column

[datlechin]

Why

Before a SQLite table rebuild that drops a column, the driver refused when any word in a trigger, view, CHECK or index matched the column name. So a column named end could not be dropped past a trigger's END, and a same-named column in another table blocked it. Real breakage also got through: a trigger that lists the column (INSERT columns, UPDATE SET, UPDATE OF) or inserts by position failed later when it fired.

Change

• A rebuild lets SQLite decide: after the views and triggers are made again, an ALTER TABLE re-checks the schema, and a view or trigger that reads the dropped column rolls the save back with SQLite's own message naming it. legacy_alter_table is pinned OFF.
• Column lists SQLite never checks (INSERT columns, UPDATE SET, upsert, UPDATE OF) are read exactly and refused before Save, for the rebuild and the native DROP COLUMN.
• The table's own CHECK constraints, generated columns and indexes are read with SQLite's grammar, so function names, collations, CAST types and keywords no longer count as columns.
• A trigger that inserts into the table by position refuses adding a column, or dropping one that isn't generated, naming the trigger.

Verified

Every new test fails on the previous driver code and passes now. Core and SQLite driver tests (743) and the full gtk suite pass.

Replaced on purpose: dropping_a_column_a_view_uses_is_refused_in_a_rebuild and dropping_a_column_a_view_needs_is_refused_before_anything_runs. A view that reads a dropped column is now refused at Save with a rollback (a_dropped_column_something_reads_rolls_the_rebuild_back).