Support filling an SChunk with a specific value. This fixes #250

[martaiborra]

No description provided.

[FrancescAlted]

Well done! I just have a couple a suggestions. If possible, please also make sure that the different NDArray constructors use this new way of building SChunks.

Also, it would be nice if you can attach a message in the PR with the improvements in speed for e.g. blosc2.zeros() and blosc2.full()

[martaiborra]

Ran a bench comparing 3 ways of filling a SChunk. The output is the following:

Time for filling the schunk with `data` argument in the constructor: 0.011s
Time for filling the schunk without passing directly the value: 0.009s
Time for filling the schunk passing directly the value to `fill_special`: 0.006s

[FrancescAlted]

Great work!

FWIW, I simplified the benchmark (essentially letting the chunksize to be computed automatically) and pushed the size of the schunk (or better, the ancillary NumPy array) towards the limits of my machine (Intel i13900K, 64 GB RAM), and here are the results:

(python-blosc2) faltet@beast:~/blosc/python-blosc2$ python bench/fill_special.py
Filling a SChunk with 2.0 Gelements of dtype=dtype('float64')
Time with `data` argument in constructor:            1553.859 ms
Time without passing directly the value:             1580.115 ms
Time passing directly the value to `fill_special`:      0.072 ms
(python-blosc2) faltet@beast:~/blosc/python-blosc2$ python bench/fill_special.py pers
Filling a SChunk with 2.0 Gelements of dtype=dtype('float64')
Writing output to bench_fill_special.b2frame...
Time with `data` argument in constructor:            1556.498 ms
Time without passing directly the value:             1653.402 ms
Time passing directly the value to `fill_special`:     47.593 ms