More robust check for compiler openmp support and enable lto

[ljgray]

No description provided.

[ljgray]

Checking for the specific cpu is annoying for macs - only very recent versions of clang support -march=native, and there's no equivalent to autodetect, so we have to check for each different M-series CPU. They also aren't all compatible with each other.

We also try to use apple Accelerate where openmp isn't available. There's an edge case where blas/lapack calls in a prange cython loop results in Accelerate spawning it's own threads and slowing things down. I'm sure there's a better way around this, but for now I don't think we care too much. I'm open to other suggestions here, though.

Finally, I guess it doesn't really matter if we pass all these flags to both the compiler and the linker, so for simplicity I've switched to using a single list.

[ljgray]

The old format is deprecated, so we need to update this everywhere. I'm using 2.0 or later here, but if there's an argument to use 2.0 only I'm fine with that as well

[ketiltrout]

2.0 or later is the right one to use.

[ketiltrout]

2.0 or later is the right one to use.