Improve definition of dependency table column names and dtypes

[hagenw]

As stated in #416 (review), our current definition of the column names and dtypes of the dependency table is done not in a nice way:

audb/audb/core/define.py

Lines 23 to 64 in 44df511

	class DependField:
	r"""Fields stored in dependency table."""
	ARCHIVE = 0
	BIT_DEPTH = 1
	CHANNELS = 2
	CHECKSUM = 3
	DURATION = 4
	FORMAT = 5
	REMOVED = 6
	SAMPLING_RATE = 7
	TYPE = 8
	VERSION = 9
	DEPEND_FIELD_NAMES = {
	DependField.ARCHIVE: "archive",
	DependField.BIT_DEPTH: "bit_depth",
	DependField.CHANNELS: "channels",
	DependField.CHECKSUM: "checksum",
	DependField.DURATION: "duration",
	DependField.FORMAT: "format",
	DependField.REMOVED: "removed",
	DependField.SAMPLING_RATE: "sampling_rate",
	DependField.TYPE: "type",
	DependField.VERSION: "version",
	}
	DEPEND_FIELD_DTYPES = {
	DependField.ARCHIVE: "string[pyarrow]",
	DependField.BIT_DEPTH: "int32[pyarrow]",
	DependField.CHANNELS: "int32[pyarrow]",
	DependField.CHECKSUM: "string[pyarrow]",
	DependField.DURATION: "float64[pyarrow]",
	DependField.FORMAT: "string[pyarrow]",
	DependField.REMOVED: "int32[pyarrow]",
	DependField.SAMPLING_RATE: "int32[pyarrow]",
	DependField.TYPE: "int32[pyarrow]",
	DependField.VERSION: "string[pyarrow]",
	}
	DEPEND_INDEX_DTYPE = "object"

I guess, the main motivation behind the current structure was to make it similar to how DependType and DEPEND_TYPE_NAMES is defined. But I would rather also change those definition, than staying with the current solution.

One possible solution would be to create a single dictionary, that stores a mapping from column name to dtype.
The only thing, not directly fitting is the DEPEND_INDEX_DTYPE, as the index is a special column, and we need to see, if we can add it to the dictionary or not.