Cymr

This version fixes an issue with how edge cases were handled when calculating log likelihood using `cymr.cmr.CMR.likelihood`. This could cause problems with data where participants sometimes recalled all items in a list.

This version reorganizes the CMR API to make it easier to use. Internally, the modules have also been reorganized to separate model-agnostic code (the general Parameters class) from model-specific code (CMR-specific parameters). This should make it easier to add support for other models in the future.

API changes:
* Saving and loading model patterns is now done using the CMR module (use `cymr.cmr.load_patterns` and `cymr.cmr.save_patterns` instead of `cymr.network.load_patterns` and `cymr.network.save_patterns`).
* CMR parameter configuration is now specified by the model-specific `cymr.cmr.CMRParameters` class instead of the generic `cymr.parameters.Parameters` class.
* Reading CMR configuration files is now done using `cymr.cmr.read_config` instead of `cymr.parameters.read_json`.

This version updates the API for parameter sweeps to match other model methods.

This patch fixes the version number to allow updating PyPI.

This version adds a system for managing general configuration variables and adds experimental support for filtering recalls using a mechanism similar to the CMR2 model.

Summary of changes:
* Parameter definitions now have support for general configuration variables. These are intended for settings that cause categorical shifts in model behavior, as opposed to graded changes that are specified using standard parameters.
* Support for changing the allowed scope of recall in CMR simulations. The scope, which may be either `list` or `pool`, determines which items are represented in the network and may compete for recall.
* Draft support in CMR for an optional recall filtering mechanism. The context associated with each candidate item is compared to the current state of context. The level of context match determines the probability of a recall being judged as matching.
* Likelihood evaluation now returns statistics for each subject separately within a DataFrame.

This version fixes a problem in CMR.generate when using patterns loaded from an hdf5 file.