[Feature] Cleanup leftover "temporary" tables

## Problem

The adapter currently implements temporary tables as actual tables.
They are not deleted after being used, that's why they start clogging available resources.

## Feature request

When a special naming convention is followed, for example prefixed by `tmp_`, left over tables can be easily identified and dropped. This feature request is to identify a suitable injection point where "old enough" temporary tables could be dropped.

## Alternatives

Manual/scheduled clean-up outside of dbt.

## References
- https://community.cratedb.com/t/cratedb-incremental-model-materialization-not-working-with-dbt-core/2010

Start by tracing how the dbt adapter implements temporary tables and how incremental model materialization completes. Identify where table age and the `tmp_` naming convention could be evaluated, then determine the cleanup policy and injection point. Done should include an agreed mechanism for safely removing old temporary tables, with coverage for the relevant materialization flow.

Written by the indexing model from the issue text.