Update missense badness for training/validation/testing

[lw453]

This PR adds changes to allow the missense badness models to be generated with variant data in the following sets of transcripts: all training transcripts, all test transcripts, or k-folds of the training transcript set. These transcript sets were defined in the notebook: gs://regional_missense_constraint/notebooks/train_test_split_transcripts.ipynb

Major changes:

• Added two reference tables for ease of working with transcripts to reference_data.py: transcript_ref and transcript_cds
• Added a utility function to filter a VEP context table to coding sites in a given set of transcripts
• Added resource paths for the training, validation, and test sets of transcripts generated in gs://regional_missense_constraint/notebooks/train_test_split_transcripts.ipynb
• Added a constant for the number of folds in the training set
• Added capability to create and write out the missense badness models and associated resources/temporary tables on the aforementioned sets of transcripts-

Minor changes:

• Updated CURRENT_FREEZE to RMC freeze 5
• Updated a couple of resource paths to use existing variables for path prefixes
• Updated format of section headings in generic.py
• Fixed a bug introduced in https://github.com/broadinstitute/regional_missense_constraint/pull/284 that caused reading and writing from the same path

Analogous changes to MPC-related code will be forthcoming in another PR.