SELECT on a UC-attached table with a collated STRING column fails

## Summary

Attaching a Unity Catalog catalog via the `uc_catalog` extension and then
querying any table that has a `STRING COLLATE UTF8_BINARY` column fails with:

```
Not implemented Error: Tried to fallback to unknown type for 'string collate UTF8_BINARY'
```

Collated `STRING` columns are the current default for string columns created
by recent Databricks Runtime versions, so this affects essentially any table
created on a modern Databricks Runtime with Unity Catalog enabled — not an
edge case.

## Environment

Reproduced identically on two DuckDB versions:

- DuckDB `v1.5.5 (Variegata) d8cdaa33fd` (current, confirmed 2026-09-07)
- `unity_catalog` extension: version `fd85147`, installed from the
`core` repository (`REPOSITORY` install mode)
- `delta` extension: version `45c4087`, installed from the `core`
repository
- DuckDB `v2.1.0-alpha40575 (Unknown Version) 154c2c8d5f` (initial repro,
2026-09-03)
- `unity_catalog` extension: version `b593157d11`, installed from the
`core` repository
- `delta` extension: version `10a49159f8`, installed from the `core`
repository
- OS/arch: macOS, `osx_arm64`
- Unity Catalog server: a Databricks workspace's built-in Unity Catalog
metastore.

## How I connected

Using the `uc_catalog` extension against a live Databricks Unity Catalog
metastore (not the OSS Unity Catalog server), with a Databricks personal
access token obtained via the Databricks CLI:

```sql
INSTALL uc_catalog;
LOAD uc_catalog;
INSTALL delta;
LOAD delta;

CREATE SECRET uc_secret (
TYPE uc,
TOKEN '',
ENDPOINT 'https://XXX.databricks.com',
AWS_REGION 'us-east-1'
);

ATTACH 'users' AS users (TYPE uc_catalog, SECRET uc_secret, DEFAULT_SCHEMA 'default');

SELECT COUNT(*) FROM users.XXX.XXX;
```

The `ATTACH` itself succeeds. The failure happens on the subsequent
`SELECT` against a table containing a collated `STRING` column.

## Error

```
Not implemented Error: Tried to fallback to unknown type for 'string collate UTF8_BINARY'
```

Reproduced consistently against multiple tables in the same catalog/schema,
all created by a recent Databricks Runtime (where collation is now the
default for `STRING` columns)

## Root cause (isolated)

Reading the *same* table's underlying data directly off S3 with the `delta`
extension's `delta_scan()` — bypassing the `unity_catalog` extension
entirely — works with no error and returns the expected row count:

```sql
INSTALL delta;
LOAD delta;
CREATE SECRET s3_secret (TYPE s3, PROVIDER credential_chain, ...);
SELECT COUNT(*) FROM delta_scan('s3:///.../tables//');
-- 606
```

This shows the `delta` extension itself can read and decode the underlying
Parquet/Delta files for this table without issue. So the failure is
specific to the `unity_catalog` extension's `ATTACH` layer: when it maps
Unity Catalog's column type metadata to a DuckDB logical type, it doesn't
know how to represent (or fall back for) the collated-string type
(`STRING COLLATE UTF8_BINARY`), and errors out before the `delta` extension
ever gets a chance to scan the file.

Start at the unity_catalog extension's ATTACH layer and its Unity Catalog column-type metadata mapping, focusing on handling for `STRING COLLATE UTF8_BINARY`. Use the supplied ATTACH and SELECT reproduction, with `delta_scan()` as the working comparison; done means SELECT succeeds and returns the expected row count for tables with collated STRING columns.

Written by the indexing model from the issue text.