Add Blosc2 filter

[oscargm98]

This PR implements Blosc2 filter support. For more details see: RELEASE_NOTES.rst

[t20100]

Some early feedbacks from starting integration of the blosc2 filter in hdf5plugin (silx-kit/hdf5plugin#201) and to answer silx-kit/hdf5plugin#201 (comment):

• For the default number of threads, I would use 1 and allow the users to change it if needed (through an env var?). The issue I see with setting it by default to something else than 1 is that it's not a good default when using multiprocessing.
• For the filter cd_values, I've just checked and it seems possible to pass an arbitrary number of integers. What about leveraging this and allow to set a pipeline of filters rather than using the same parameters as for blosc1? And for example passing (0, 0, 0, 0, compression, compression level, filter1, filter1_param, filter2, filter2_param, varying number...). It probably does not cover all cases, but would already enable using delta or trunc_prec + bitshuffle. Else this is only available through direct chunk write

[FrancescAlted]

Some early feedbacks from starting integration of the blosc2 filter in hdf5plugin (silx-kit/hdf5plugin#201) and to answer silx-kit/hdf5plugin#201 (comment):

• For the default number of threads, I would use 1 and allow the users to change it if needed (through an env var?). The issue I see with setting it by default to something else than 1 is that it's not a good default when using multiprocessing.

That's a good suggestion. We have just implemented support for BLOSC_* env vars (in particular BLOSC_NTHREADS) for the super-chunk API in C-Blosc2 2.5.0, and are working towards releasing Python-Blosc2 including C-Blosc2 2.5.0. Once this would be there, we can proceed to make the default number of threads to 1 and control that via BLOSC_NTHREADS.

• For the filter cd_values, I've just checked and it seems possible to pass an arbitrary number of integers. What about leveraging this and allow to set a pipeline of filters rather than using the same parameters as for blosc1? And for example passing (0, 0, 0, 0, compression, compression level, filter1, filter1_param, filter2, filter2_param, varying number...). It probably does not cover all cases, but would already enable using delta or trunc_prec + bitshuffle. Else this is only available through direct chunk write.

Good to know and agreed. Let's start with something simple first and let's see how we can accommodate more parameters in the future.

[FrancescAlted]

We think this is mostly ready to be merged. If nobody opposes, we will merge this and start the release process during the next week.

[avalentino]

It is mostly OK for me.
Just a minor comment.