Updated c-blosc to commit 9dc93b1 and zstd to v1.5.2

[t20100]

This PR updates c-blosc to commit 9dc93b1 initially to fix macos universal2 build.
It also updates zstd to v1.5.2 which includes a piece of x86_64 assembly code (that looks to work by passing it as extra_objects to the Extensions).
This still needs testing on architecture different from x86_64.

closes #186

[t20100]

The assembly code is only enabled under linux and macos and explicitly disabled otherwise.
There is further checks with C macros to check if it is actually there.

Tested also on Power9 and macos ARM.

One issue may be for generic builds (manylinux, conda): This assembly code uses BMI2 instructions which are apparently available starting from Haswell Intel processors and Excavator AMD processors, so available since about 10 years.

We can:

• Disable it all the time
• Add (yet another) env. var. to configure the build (e.g., HDF5PLUGIN_BMI2).

[t20100]

I added an env. var. HDF5PLUGIN_BMI2=True|False (default True) to allow disabling this.

Ready for review.

[vasole]

LGTM. I like the conservative approach concerning assembler code and windows...

[t20100]

I like the conservative approach concerning assembler code and windows

If you know how to handle assembly code in build through setuptools on Windows, I'd take it.

[t20100]

From a quick look (I haven't tested anything), Visual Studio does not seems to support assembly code by default.
From what I've seen, one need to install an extra tool like nasm to compile the assembly code first.
With gcc and clang, one can pass the .S file directly at link time: thus the use of extra_objects in this PR and that I disabled it elsewhere including Windows.
We can consider enabling it for Windows when it is not using Visual Studio, but it is really a corner case and not worth the effort to me.