Improve the error handling in the backward and forward convergence

[xiki-tempula]

In the backward and forward convergence, for the initial set of points, which uses for example the first 10% of the date, it could be the cases where due to the fact that there are not many data points, so the overlap is pretty bad, which gives terrible statistical error.
This PR is build on #357.
The PR allow the user to specify a error tolerance. If the statistical error is too bad, use the bootstrap error instead choderalab/pymbar#519.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.83%. Comparing base (46cc83b) to head (4ac7596).
Report is 10 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #358   +/-   ##
=======================================
  Coverage   98.83%   98.83%           
=======================================
  Files          28       28           
  Lines        1890     1895    +5     
  Branches      409      407    -2     
=======================================
+ Hits         1868     1873    +5     
  Misses          2        2           
  Partials       20       20           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[orbeckst]

@xiki-tempula you'll have to manually resolve these conflicts

[orbeckst]

Did you remove the speed-up for the convergence estimate (ie using the bar estimate)? If so update CHANGES (and docs where necessary).