Megatron

Bug Fixes
- Node names are fixed, by delegating the responsiblity of naming solely to the node itself.

Changes
- Add Flatten layer.
- Add Filter layer.
- Give layers a name attribute rather than manipulate the class name.
- Make more descriptive error messages; when a node fails, identify which node it is.
- Make use of progress bar for full dataset fits as well as generator fits, for Keras models.

Bug Fixes
- Fix dataset loaders, which previously only threw an error.

Changes
- Add Slice layer.
- Re-arrange nodes module.
- Add exploration mode to the pipeline for EDA.
- Separate out paths for transformation, evaluation, and exploration within the pipeline to avoid unnnecessary computation.
- Update visualization module to include exploration nodes in their own colour.

Changes
- Add Normalization layer.
- Convert TimeSeries layer to Stateful so it can work across batches without loss of data.
- Add traversal method to Nodes to be able to navigate between them.
- Add 'anon' to label of Lambda layers that are anonymous functions.

Changes
- Add ability to overwrite storage for a pipeline name/version.
- When executing eagerly, creating a pipeline destroys stored data and metadata.

Bug Fixes
- Fix pipeline saving by re-structuring usage of dynamic classes in wrappers.

Changes
- Merge metrics module into layers.

Bug Fixes
- Remove extra argument in CSVData.
- Fix storage read method.
- Re-work unused data pruning logic.