A naïve implementation of the expression that feeds one value at a time into an accumulator works but likely can be improved perf-wise. We should consider:

1. sending the whole input at once rather than each value one by one
2. not using t-digest for small inputs