Remove struct implicit casting when names mismatch and prevent table initialization using ROW

[maiadegraaf]

This PR is a slightly different implementation of #8854 but does the same thing but more succinctly.
Solves issues #3884, #7303, #7765, and #8301.

---

Structs can no longer be inserted unless the names match.
The following will now throw an error:

CREATE TABLE t1 (s STRUCT(d DOUBLE, v VARCHAR)); 
INSERT INTO t1 VALUES ({i: 20, v: 'foobar'}), ({i: 20, v: 50});

The ROW function is still supported and unnamed structs are properly auto-casted to any struct.

---

Additionally tests can no longer be initialized using the ROW function, the following will now throw an exception:

CREATE TABLE t1 AS (
	SELECT ARRAY[
		(1, 'x'),
		(2, 'y'),
		(4, 's')
	] AS list
);

And should be rewritten as the following:

CREATE TABLE t1 (list STRUCT(a INT, b VARCHAR)[]);
INSERT INTO t1 VALUES (ARRAY[(1, 'x'), (2, 'y'), (4, 's')]);

[Tishj]

To not kill the usability of CREATE TABLE AS, it's also possible to use

statement ok
CREATE TABLE t1 AS (
	SELECT ARRAY[
		(1, 'x'),
		(2, 'y'),
		(4, 's')
	]::STRUCT(a int, b varchar)[] AS list
);

So it doesn't have to be a two step process

[taniabogatsch]

LGTM! I only have two minor questions.

[taniabogatsch]

Why are these changes included in this PR?

[maiadegraaf]

Because spatial tests were failing I had to update the, and therefore add the most recent version of spatial to the CI run, but @Maxxen had also created new functions in spatial so those had to be added too.

[taniabogatsch]

Do you know if this is still the case? I think that we are using default names for the struct fields, as this is not the ROW function.

[Tishj]

import duckdb
from duckdb.value.constant import StructValue

duckdb.execute("""
	create table tbl (a STRUCT(a BIGINT, b VARCHAR, c BOOL))
""")

# Equivalent to ROW(42, 'bla', True)
duckdb.execute("insert into tbl select $col_a", {
	'col_a': StructValue((42, 'bla', True), [int, str, bool])
})

res = duckdb.table("tbl").fetchall()

print(res)

Currently this passes, because we don't check for name equality, but after this PR this no longer works because the provided struct value is considered "named"

---

Internally this calls struct_type(children) where children here is [int, str, bool]

[Mytherin]

Thanks for the PR! Looks good - some comments:

[Mytherin]

I would say that () should always create an unnamed struct. That's the behavior in Postgres.

Perhaps we can also change the ToString of unnamed structs to leave out the name entirely, although that might be better done in a future PR.

[Tishj]

I also thought about the ToString modification, but that would break roundtripping of the type string, which is something to consider

[Mytherin]

Thanks! LGTM