S3 multi-part cleanup in destructor path

[ihnorton]

The purpose of this issue is to improve the following sequence which can cause network calls in the tiledb::StorageManager destructor path:

• ~StorageManager
• calls VFS::terminate
• calls S3::disconnect

Merging https://github.com/TileDB-Inc/TileDB/pull/1325 will mitigate the slowdown reported in https://github.com/TileDB-Inc/TileDB/issues/1324 in the common case when there are no unfinished multipart uploads.

However, the clean up code in the ~StorageManager destructor path will still be reachable in at least the following two situations co-occurring with incomplete uploads:

• per @joe-maley, upon error occurring in the tiledb_vfs_write path, which is a user-facing API.
• upon error during a global order write