add tl.math.tanh instead of tl.libdevice.tanh

[michaelfeil]

Seems like tanh moved to a

see triton/language/math.py-L1238

@core.extern
def tanh(arg0, _builder=None):
    return core.extern_elementwise(
        "libdevice", libdevice_path(), [arg0], {
            (core.dtype("fp32"), ): ("__nv_tanhf", core.dtype("fp32")),
            (core.dtype("fp64"), ): ("__nv_tanh", core.dtype("fp64")),
        }, is_pure=True, _builder=_builder)

[michaelfeil]

@pommedeterresautee FYI, unit tests seem to pass. What do you think about this PR?

[michaelfeil]

@pommedeterresautee friendly reminder!