r.mapcalc: remove static variables from mode and median to fix parallelization

[petrasovaa]

This is an attempt to get rid of static variables from the r.mapcalc functions for mode and median. The static variables were there I assume to not allocate for every row, but it complicates parallelization, see #5742. Since the size of the array is typically very small (the number of arguments of the median() or mode() function, can be raster or number), it will use array on stack and allocate only for larger sizes.

I am not a C expert, but hopefully this makes sense.

[wenzeslaus]

I consider this a good tradeoff. It may be slower, but it will reliably work with parallelization. If it is too slow, we can optimize by creating a context structure which would be unique for each thread.

The compilation is broken given the more strict requirement in the CI on C standard.

[nilason]

I think this makes sense. It might even be faster for cases within the threshold (using heap allocated memory instead of realloc), that being without the additional positive prospects of parallelisation...

[wenzeslaus]

Thanks.

As for the f-strings, I really meant just the short ones in the setup, the remaining ones are matter of taste, but may trigger some linters.