by preprocessing the input `Rcpp::StringMatrix` into a
`xt::tensor<std::array<char, 32>>` to avoid the limitation of strings
not being supported as a valid data type for `xt::rtensor`

These are required for the new probabilistic metrics.

eventhough the conditional masking is not yet accessible in `evalhyd-r`,
it impacts on the number of axes for *t_msk* (from 3D to 4D to add
an axis for leadtimes), so it needed to use the latest version of
`evalhyd`.

which includes a temporal masking functionality for deterministic
evaluation

version which includes the addition of the "sites" axis to
dimensions of thresholds and masks

which adds a pre-processing functionality to `evald` to
apply transform functions to streamflow observations and predictions
prior to the computation of the metrics (new parameters *transform*,
*exponent*, and *power* are included in the Python API), and which
includes a few more determinist metrics (i.e. RMSE, KGE, KGEPRIME)

which adds the temporal masking functionality, and
the sites/leadtimes dimensions to the inputs

i.e. where metrics is put as last positional parameter

Since R does not easily support subpackages/submodules, the
deterministic `evaluate` function is renamed `evald`, and the newly
created probabilistic one is named `evalp`. In order to make compile
the source files (*.cpp) in the probabilistic part of evalhyd, the
Makevars files needed to do a bit more work to find the source files
in the evalhyd dependency and place the object files (*.o) in the *src*
directory for the R package building to find them.

The new API of `evalhyd` is centred around two `evaluate` functions,
one for deterministic evaluation and one for probabilistic evaluation.
The probabilitist side of things is not header-only anymore, and so
additional C++ source files needed to be pointed, which is not working
as of yet. So this version of the R-bindings only exposes the
determinist `evaluate` function for now. The example Rmarkdown is also
updated to reflect those changes.