Some of the tokens forming a `GENERATED ALWAYS` column constraint
sometimes end up in the column type. I think this is inconsistent
with the syntax diagrams (when these tokens are interpreted as
`type-name` they cannot also be interpreted as constraint syntax)
and internally inconsistent (e.g., this happens with anonymous
column constraints, but does not happen if the column constraint
is named):
```
sqlite> create table t1(id int, x int generated always as (true));
sqlite> create view v1 as select * from t1;
sqlite> create table t2(id int, x int constraint c generated always as (true));
sqlite> create view v2 as select * from t2;
sqlite> select * from PRAGMA_TABLE_XINFO('t1');
cid name type notnull dflt_value pk hidden
--- ---- -------------------- ------- ---------- -- ------
0 id int 0 0 0
1 x int generated always 0 0 2
sqlite> select * from PRAGMA_TABLE_XINFO('t2');
cid name type notnull dflt_value pk hidden
--- ---- ---- ------- ---------- -- ------
0 id int 0 0 0
1 x int 0 0 2
```