Add audb.stream() and audb.DatabaseIterator (#448)

* Add map argument to audb.load_table()

* Improve docstring example

* Add audb.stream()

* Add more ideas

* Create first working version of stream

* Move to extra file

* Use extra class for parquet

* Add solution for csv files

* Add shuffle support for CSV

* Finish first implementation

* Add string representation

* Adjust docstring

* Add to documentation

* Fix raises section of docs

* Use DatabaseIterator name, shuffle in example

* Fix map

* Simplify docstring example

* Inherit DatabaseIterator from audformat.Database

* Improve example

* Fix type annotations of __next__ + __iter__

* Fix storing in flavor cache folder

* Use __iter__ form audformat.Database

* Revert "Use __iter__ form audformat.Database"

This reverts commit 2bfc800.

* Fix buffering

* Add first tests

* Add tests and fix bugs

* Extend docstring of audb.stream()

* Add section to usage documentation

* DEBUG Windows

* DEBUG Windows

* Try chunks for CSV reader

* Restructure Iterator classes

* Try to fix doctest under Windows

* Set default value of buffer_size to 100_000

* Update docs/load.rst

Co-authored-by: ChristianGeng <christian.c.geng@gmail.com>

* Use **kwargs to simplify code

* Remove __init__ function in inherited Iterators

* Make surfe persistent repository is cleaned

* Make DatabaseIterator abstract class

---------

Co-authored-by: ChristianGeng <christian.c.geng@gmail.com>

audb/__init__.py

@@ -19,6 +19,8 @@
 from audb.core.load_to import load_to
 from audb.core.publish import publish
 from audb.core.repository import Repository
+from audb.core.stream import DatabaseIterator
+from audb.core.stream import stream
 
 
 __all__ = []

audb/core/load.py

@@ -25,7 +25,7 @@
 from audb.core.utils import lookup_backend
 
 
-CachedVersions = typing.Sequence[typing.Tuple[audeer.StrictVersion, str, Dependencies],]
+CachedVersions = typing.Sequence[typing.Tuple[audeer.StrictVersion, str, Dependencies]]
 
 
 def _cached_versions(
@@ -805,7 +805,28 @@ def _misc_tables_used_in_scheme(
         if scheme.uses_table:
             misc_tables_used_in_scheme.append(scheme.labels)
 
-    return list(set(misc_tables_used_in_scheme))
+    return audeer.unique(misc_tables_used_in_scheme)
+
+
+def _misc_tables_used_in_table(
+    table: audformat.Table,
+) -> typing.List[str]:
+    r"""List of misc tables that are used inside schemes of a table.
+
+    Args:
+        table: table object
+
+    Returns:
+        unique list of misc tables used in schemes of the table
+
+    """
+    misc_tables_used_in_table = []
+    for column_id, column in table.columns.items():
+        if column.scheme_id is not None:
+            scheme = table.db.schemes[column.scheme_id]
+            if scheme.uses_table:
+                misc_tables_used_in_table.append(scheme.labels)
+    return audeer.unique(misc_tables_used_in_table)
 
 
 def _missing_files(