Runtime check and optional no_std

[lisicky]

Despite that the previous PR #39 fixes compilation without x86 SIMDs flags, there is the case when a user of cryptoxide compiles for x86 and tries to execute code with cryptoxide can get an error SIGILL, because the code branch below works only in compile time.

By this PR I suggest adding a runtime SIMD support check and making cryptoxide std friendly because the is_x86_feature_detected is only available in the std. But I left the possibility to build cryptoxide as no_std by disabling default features. I can make an update for README.MD with instructions on how it can be done. That no_std manner I got from https://github.com/marshallpierce/rust-base64 and https://github.com/iqlusioninc/crates/tree/main/bip32 repos.

cryptoxide/src/hashing/blake2/mod.rs

Lines 113 to 120 in 9d22605

	#[cfg(target_feature = "avx2")]
	const HAS_AVX2: bool = true;
	#[cfg(not(target_feature = "avx2"))]
	const HAS_AVX2: bool = false;
	if HAS_AVX2 {
	return avx2::compress_b(&mut self.h, &mut self.t, buf, last);
	}

[vincenthz]

in term of runtime check I think it's good, but not adding back std (or adding dependencies) is a stronger priority of mine for cryptoxide. Likely I imagine, adding a cpu.rs module with some inline assembly / cpuid checks. we need a very small subset of std::is_x86_feature_detected