Layer base get out data, use dim tags and correct batch

[albertz]

No description provided.

[albertz]

test_attention_no_encoder_dependency fails:

returnn: INFO: TensorFlow exception: assertion failed: [Data(name=\'prev:accum_att_weights_output\', batch_shape_meta=[B&Beam{\'output/prev:output\'}(4),F|F\'att_energy:feature-dense\'(1),T|\'enc time\'[B]])] [shape] [8 1 5] [dyn_size[2]] [5 5] [.shape] [2] [-> invalid shape(dyn_size[2]) or invalid batch dim] [8]
	 [[node output/rec/prev__accum_att_weights/runtime_sanity_check/Assert_3/Assert (defined at home/runner/work/returnn/returnn/returnn/tf/util/data.py:1378) ]]

This should have been fixed by better dim tag batch handling and automatic beam expansion (#613).

[albertz]

Ok, after a rebase, still this:

InvalidArgumentError: assertion failed: [Data(name=\'prev:accum_att_weights_output\', batch_shape_meta=[B&Beam{\'output/prev:output\'}(4),F|F\'att_energy:feature-dense\'(1),T|\'enc time\'[B]])] [shape] [8 1 5] [dyn_size[2]] [5 5] [.shape] [2] [-> invalid shape(dyn_size[2]) or invalid batch dim] [8]
	 [[{{node output/rec/prev__accum_att_weights/runtime_sanity_check/Assert_3/Assert}}]]

Which still looks like #613.

[albertz]

Still the same after rebase:

InvalidArgumentError: assertion failed: [Data(name=\'prev:accum_att_weights_output\', batch_shape_meta=[B&Beam{\'output/prev:output\'}(4),F|F\'att_energy:feature-dense\'(1),T|\'enc time\'[B]])] [shape] [8 1 5] [dyn_size[2]] [5 5] [.shape] [2] [-> invalid shape(dyn_size[2]) or invalid batch dim] [8]
	 [[{{node output/rec/prev__accum_att_weights/runtime_sanity_check/Assert_3/Assert}}]]

[albertz]

This was part of #613 now.