Updates the logging initialisation so that, when called with -q flag,
the log file is still written with INFO level log messages (only the
console output is supressed).

Also fixes big in generate_initial_p0

If an initialisation step is callled, the sampler runs the simulations
for nsteps and selects the point with the highest lnprobability from
which to restart the simulations. Previously, only the last step was
used. In this update all steps are used and a message is printed to
allow the user to determine if the selected point is appropriate.

1) Colors the output depending on if it is burn-in or production
2) Adds another panel with the twoF values of all walkers (also colored
by bunr-in or prod)
3) Also adds more logging to writing par files

1) Fixes error in glitch logp ts check
2) Adds ts check to logp to avoid raising ValueError when one of the
   glitch times is before the tstart
3) Fixes bug in BSGL so that it can be used with glitch searches
4) Generates the output directory before attempting to write log file

- Adds mismatch spacings
- Adds maxNlocator option

Note: this will break the tests which have not yet been updated

1) Adds check if the cached data will be used to avoid loading the
search in the grid search
2) Adds better handling of errors in the par read
3) Adds ability to add 2d plots to an existing figure

Adds a switch to calculate (or optimize over) BSGL rather than 2F. This
should help to quiten noise events which exist only in one detector.

- Adds ability to truncate data to a specified period
- Removes the write-after option from the grid search
- Adds option to just return the data array from the grid search

It seems that , due to the choice of transient windows, occasionally
the calculated Fstat is +/- inf or a nan. This behaviour is not
understood. To address this, we apply a simple check and if it is an
undesirable value, return 0. This will inherently bias the chains away
from regions of parameter space in which the output breaks.

1) Improves the output
2) Fixes a bug in which the while loop got stcuk attempting to rename
the glitch parameters (when ng > 1)

If the initial p0 values are -np.inf due to the prior, it will iterate
through the other p0 values to replace the duff values. The replacement
takes another randomly selected p0 value and shifts it round by a
fractional gaussian with std=1e-10

1) Add rcparams to prevent all use of offsets
2) Fixes error when nglitch > 1 in which the summary did not show all
values
3) Fixes error in which the max_twoF used incorrect indexes

Also various improvements to logging commands

1) Use calculate_thetas to precalculate all thetas and then iterate
through to calculate twoF values. This will be important when we want to
allow theta to run both from and two theta0
2) Add diagnostic information on the mean acceptance fraction