This PR fixes https://github.com/duckdblabs/duckdb-internal/issues/1371

As the title says, this PR adds a setting to control maximum swap space.

Using the temp_directory setting, a directory can be assigned to use by DuckDB to offload temporary intermediate data, to evict buffers that can not be processed yet but are taking up room required by other buffers, we store them on disk.

"Disk is full" issue

The problem this solves is that this can grow unbounded, in the worst case eventually flooding the entire disk, making it unusable.
To prevent this issue, we will instead abort the query if the max_temp_directory_size is reached

max_temp_directory_size

The setting defaults to 0 when temp_directory is not set.

If temp_directory is set to an existing directory, it will default to the available disk space on that drive.
If temp_directory is set to a non-existent directory, it will default to 0. When the directory gets created by us, we'll update it to the available disk space on that drive.

When the setting is set explicitly by the user, it will not be influenced by the temp_directory setting at all.
(see test/sql/storage/max_swap_space.test for examples of this)
When the value is "none", "null" or the value starts with - then the swap limit is set to unlimited.

Future work

Maybe in the future, instead of aborting the query, we can let the task block instead, to be rescheduled later when enough disk space is available
(realizing this might be a more general use for blocking tasks, rather than assigning a special event that is responsible for unblocking events, maybe we want to have a generic system that periodically checks a callback to detect if a task should be unblocked)

Other changes

FileSystem now has a static method GetAvailableDiskSpace which takes a path and retrieves the disk space info from it.

DatabaseInstance::Initialize no longer modifies the DBConfig passed in by the user, these modifications are now done directly to the internal config inside DatabaseInstance::Configure