Add NEON float16 multi-vectors to native aliases

[stellar-aria]

Half-precision multi-vector types are available on the VFPv3_fp16 and VFPv4 architectures for ARMv7

[mr-c]

Thank you @stellar-aria for this PR. I see nothing objectionable, but it would be better if these new types were used elsewhere in SIMDe, so we could test that the various compilers work with them properly.

Do you have any suggestions?

[stellar-aria]

The only functions that really use the types are the interleaved load/store intrinsics (i.e.
vld[n]_f16, vld[n]_dup_f16, vld[n]_lane_f16 and vst[n]_f16, where n is the stride between 2 and 4)

How are the other native aliases used/tested? Just attempting to have functions declared that used the types (with the native aliases enabled) was enough to get compiler errors for me.