cpu_features improvements

[metalefty]

Changes proposed in this pull request:

• Update bundled cpu_features to 0.9.0
• Allow the use of cpu_features for more architectures (aarch64, s390x, etc) on Linux [1]
• Add USE_SYSTEM_CPU_FEATURES flag to use system's cpu_features instead of the bundled one
• Use system's cpu_features in FreeBSD CI (for the proof of USE_SYSTEM_CPU_FEATURES)

[1] https://github.com/google/cpu_features/tree/v0.9.0?tab=readme-ov-file#whats-supported

[metalefty]

I'll fix the rawhide CI failure.

[chipitsine]

btw, I'm not a big fan of git modules.
as we can pick it from system, i.e. USE_SYSTEM_CPU_FEATURES, can we use it as only approach ? or do we miss it on Windows ?

[chipitsine]

I'll fix the rawhide CI failure.

I'd suggest to switch to USE_SYSTEM_CPU_FEATURES where possible

[metalefty]

I made this change not to change the current structure and behaviour too much. However, I personally like to switch to the system cpu_features, like libsodium, rather than this approach. I'll raise another PR, let me some time.

I'm not sure about Windows BTW.