Adapt llvmcall()

[briederer]

Following the new conventions introduced in JuliaLang/julia#36576 I've adapted the internal llvmcalls such that basic functionality for arithmetic functions and floor/ceil is restored.

Also I am right now having a look at a good solution to add additional basic functions (like tan()) to find a solution for #5. It seems to me that this issue was originally caused by something outside the BitFloats since on my machine (using julia 1.7.1) I cannot reproduce #5 but instead simply get:

julia> tan(Float128(2.0))
ERROR: MethodError: no method matching tan(::Float128)
Closest candidates are:
  tan(::Real) at ~/.julia/v1.7.1/share/julia/base/math.jl:1218
  tan(::T) where T<:Union{Float32, Float64} at ~/.julia/v1.7.1/share/julia/base/special/trig.jl:215
  tan(::LinearAlgebra.Hermitian{var"#s858", S} where {var"#s858"<:Complex, S<:(AbstractMatrix{<:var"#s858"})}) at ~/.julia/v1.7.1/share/julia/stdlib/v1.7/LinearAlgebra/src/symmetric.jl:761
  ...
Stacktrace:
 [1] tan(x::Float128)
   @ Base.Math ./math.jl:1220
 [2] top-level scope
   @ none:1

Fixes #7 since the error there is only cause by the internal use of abs() when calling eps().