- Less bias with missing values in computeEffectiveNumObs
- New function varCor to compute unbiased variance of uncorrelated time series
- seCor now based on varCor, which reduces bias for small number of effective observations.

- Implement the Lo 2012 approximation of the distribution of the difference of two lognormally distributed random variables by a shifted lognormal distribution.
- Add Formulas to vignette of lognormalSum

- Add shift argument to moments, mode, and median to deal with shifted lognormal distribution.
- Implement sample-based Distribution function for the difference of two lognormal variables.

- rewrite documentation using roxygen2 instead of inlinedocs
- getParmsLognormForLowerAndUpper: remove argument isTransScale and provide new function getParmsLognormForLowerAndUpperLog instead
- merge vignette aggregateCorrelated to vignette lognormalSum

fix for CRAN: reformat (html-escape) doi in DESCRIPTION

Sum of lognormals: if all values are gap-filled, assume multiplicative standard deviation (sigmaStar) of the sum to be the mean of given sigmaStar. Before took it assumed to be the maximum. But this led to larger confidence bounds compared to using the normal assumption.

Estimate parameters from various statistics, such as mean and an upper quantile value, i.e. a practical maximum.

Support common case of estimating standard error in the presence of
correlations: function `seCor`

.

estimateSumLognormal by default now returns NA if there are NA values in terms. New argument na.rm = TRUE allows for previous behavior of neglecting those terms.

- Compute effective number of observations
- fixed bracketing bug in formula
- better dealing with NA at begin or end

basic setup