This commit fixes a bug in the predictive time-series where
the stochastic process for the force of infection was misaligned
with the stochastic process for the events.

Investigation showed that the high rate of cases in the UK in January
caused a significant discretisation error between the chain binomial
model and continuous time analogues.  The previous Rt calculation
made use of a continuous time approximation, and so gave biased
estimates of Rt.

The new implementation calculates Rt using expectations of the chain
binomial model in terms of transition probabilities, not rates as
previously.

CHANGES:

1. Kernel builder functions moved to `mcmc_kernel_factory.py`
2. Windowed adaptive MCMC a la STAN implemented in `inference.py`
3. Prior on beta1 tightened to improve stability.
4. Depends on `gemlib`@develop branch for tf-nightly and tfp-nightly

CHANGES

1. Most tasks touched to use netCDF4 inferencedata file.
2. Added netCDF4 metadata copy to covid.util
3. Updated model_spec.py to use xarray
4. Updated covid.ruffus_pipeline to use netCDF4 files, adding pipeline metadata
5. Updated pyproject.toml
6. Removed chunk specification in inference.py.  Defer to automatic chunking for now.

CHANGES:

1. Introduced default unit traffic flow;
2. Removed old test data from data directory.  This now contains real static data only.
3. Brought README.md up to date.
4. Removed extraneous files from root project directory.

CHANGES:

1. Replaced `read_phe_cases` with `CasesData`
2. Column-name fix in `case_exceedance`
3. Fix column heading in `summarize.infec_incidence`
4. Add Rt_exceed to `summarize.rt`
5. Add layer option to `summary_geopackage`
6. Update `ruffus_pipeline.py` to reflect above changes
7. Update `template_config.yaml` to reflect above changes

Note: in the actual analysis, the effect of December Tiers was
removed.

CHANGES:

1. New data adapter `read_challen_tier_restriction`
2. Modified `model_spec.py` to use new data adapter

CHANGES:

1. Re-introduced Tier lockdown covariates.  National lockdown covariate is set to 0 because
   we cannot identify this from the time-varying baseline force of infection.

2. Added inference on variance parameter for the temporal Gaussian process on beta.

CHANGES:

1. update date and data name in enqueue_pipeline.sh
2. update infection time initialisation in model_spec.py

CHANGES:

1. Implemented log-linear model for weekday effect in I->R hazard;
2. Updated summary.py and within_between.py to match;
3. Implemented linear random walk for new gamma0 parameter.
4. Implemented per-parameter data structure in output HDF5 file.

CHANGES:

1. Updated the model to use the new gemlib library;
2. Removed automatic deletion of results directory by prepare_config.py;
3. Parameterised `gamma` prior parameters and added to template_config.yaml;
4. Updated within_between.py to use global configuration file.