[RVV] add rvv f16 kernels for velu, vgelu, vapproxgelu, vsigmoid, vtanh

[velonica0]

Add rvv f16 kernels for velu, vgelu, vapproxgelu, vsigmoid, vtanh.

Tested on SpacemiT K3 CPU(VLEN=256 and zvfh_zvfhmin).

K3 FP16 Activation Benchmark Results

Operator	N=7680 (Scalar)	N=7680 (RVV)	N=65280 (Scalar)	N=65280 (RVV)	Speedup (Max)
f16-velu	83,954	5,601	713,689	47,433	15.0x
f16-vgelu	284,363	7,785	2,408,980	66,021	36.5x
f16-vapproxgelu	237,706	7,797	2,019,673	66,018	30.6x
f16-vsigmoid	289,988	16,620	2,465,351	141,139	17.5x
f16-vtanh	170,980	8,114	1,435,735	69,170	20.8x

Note: > 1. Speedup is calculated based on $N=65280$.
2. The scalar FP16 baseline is very slow — it either uses software emulation or widen-to-f32 with extra conversion overhead.

[ken-unger]

Thanks. BTW I have f16-sin, cos, exp done and will post a PR tomorrow.