feat(python): Improve write_database, accounting for latest adbc fixes/updates

[alexander-beedie]

Closes #12684.
Ref: #10763, #11282 (comment).

Enhancements

Updates pl.write_database and associated tests with various improvements now that some key issues (apache/arrow-adbc#1000, apache/arrow-adbc#1109) have been resolved in the ADBC 0.7 and 0.8 releases, respectively.

• Enables correct resolution of schema-qualified names (eg: schema.table_name).
• Adds ADBC engine support for if_exists = "fail" (simulated).
• Adds ADBC engine support for if_exists = "replace" (native).
• All write tests now able to run under Python 3.12.

Requires that the caller has installed at least ADBC version 0.7 for "replace" support, or 0.8 if using schema-qualified names, as these versions contain the related/required fixes (if not doing either of these things earlier versions of ADBC continue to work fine).

Note

In addition to SQLite (what we run in CI), I was able to validate these updates against a local PostgreSQL instance.

[stinodego]

This looks good 👌

I'm just wondering if we should set a minimum version for the adbc dependencies. Are there any older version that are fundamentally at odds with our current implementation?

[alexander-beedie]

I'm just wondering if we should set a minimum version for the adbc dependencies. Are there any older version that are fundamentally at odds with our current implementation?

Checked that these updates work fine with ADBC 0.5 and 0.6, but I think we should raise an error if given schema-qualified table names, as robust handling of these requires a recent ADBC version (not an issue with our implementation, but we can save the user from tripping themselves up by detecting it and directing them to upgrade ✌️)