Differences in percentiles when using --base_period with a different study period?

When calculating percentiles, icclim can use the `base_period_time_range` parameter to set the reference period. If `base_period_time_range` is not defined, the `time_range` is used.
I originally calculated the percentiles over the 1985-2014 `time_range` without `base_period_time_range` however if I calculate percentiles over the 1979-2021 `time range` with a `base_period_time_range` of 1985-2014, there are some very small differences. This plot shows the R95p difference for Jan 1986 between the two

Perhaps I'm misunderstanding how `base_period_time_range` is applied, I thought that they should be the same within 1985-2014 as the reference period is the same?

I also tested it by recalculating the percentiles over the 1985-2014 `time_range` with a `base_period_time_range` of 1985-2014 and this gave the same result as when using the 1985-2014 `time_range` without `base_period_time_range`

Start by tracing how percentile calculations apply `base_period_time_range` and `time_range`, using the reported R95p result for January 1986 as the comparison case. Reproduce both configurations and determine whether identical reference periods produce identical percentiles; done means explaining or correcting the observed differences and verifying the result.

Written by the indexing model from the issue text.