Delayedarray

- chore: Remove Python 3.8 (EOL).
- precommit: Replace docformatter with ruff's formatter.

- Coerce `numpy.generic` instances to `dtype` in the `SparseNdarray` constructor, for consistency with NumPy functions.
- Avoid cast to float when operating on uint64 indices in a `SparseNdarray`.

- Support `dtype=` and `copy=` arguments in `__array__()`, as required by NumPy v2.
- Added getters for the `RegularTick` properties, for convenience.

- Add support for more numpy functions to dense and sparse `DelayedArrays`.

- Switch to `Grid` classes to control how iteration is performed. This supercedes the previous `chunk_shape()` generic.

- Added a `buffer_size=` option to the `apply*` functions.
- Provide converter generics to SciPy classes, with methods for `SparseNdarray` and sparse `DelayedArray`s.
- Converted th `to_*_array()` functions into generics.
- Correctly handle zero-extent arrays in the `apply*` functions.
- Added methods to compute basic statistics (e.g., sum, variance) from a `SparseNdarray`.
- Added a copy method for `SparseNdarray`.
- Use block processing to compute basic statistics from a `DelayedArray`.
- Do not require Fortran storage order from `extract_dense_array()`.
- The `subset=` argument is now mandatory in `extract_*_array()` calls.
- Bugfix to respect changes in index dtype upon `SparseNdarray` concatenation.