Ability to make hyperparameters constant

[ravinkohli]

For some experiments, we may want to set certain hyperparameters to be constant. This currently can be done if we want to include components for ex, include={'trainer': ['StandardTrainer']} or for categorical hyperparameters by simply sending a single value into the value_range parameter of an update, for ex, HyperparameterSearchSpaceUpdate(node_name="network_backbone", hyperparameter="ShapedResNetBackbone:resnet_shape", value_range=['brick'], default_value='brick').

However, for integer or float hyperparameter, ConfigSpace gives an error if we want to set the lower and upper to the same value. This can be done either via another parameter such as 'make_constant' in the HyperparameterSearchSpaceUpdate or as decided earlier we can encapsulate ConfigSpace by introducing a search_space module that can handle such an update elegently

[ravinkohli]

We can also follow the technique used in https://github.com/automl/Auto-PyTorch/blob/7c95381d5345498ddbe026d3b8b8ae41264644c9/autoPyTorch/utils/config_space_hyperparameter.py