This brings the tuple and the enum closer together to avoid confusion,
like in Desbordante#209

`pybind11::object` manages reference counting, which is sometimes
unneeded. Also changes some `std::string` to `std::string_view`.

Allows to denote when multiple types can be accepted.

The files are structured this way because an algorithm may use several
tables (HyMD) or even a list of tables (Spider).
Those algorithms still need the `InputTable` type, but they will
register different options, which may become common later and will thus
be put in this directory as well.

Also includes some minor test rewrites. Makes it easier to implement
algorithms that take something other than a single table.

Uses a more robust way of getting the number of columns. Avoids the
reliance on `input_table_` being set to the table being processed now,
indirectly inherited from the old implementation. Previously, there was
a possibility that `input_table_` would not be set to the table being
processed in `PliBasedFdAlgorithm` because of its `LoadData` overload,
but the only scenario which used it was using `TypoMiner`, and that
class would always set the option correctly.