use isuruf's speedup

[jverzani]

Use @isuruf's rewrite of get_symengine_class in #285 to reduce memory allocation with get_symengine_class

This makes dispatch on BasicType nearly as performant as using ifelse statements.

A test case (relevant to #279)

using SymEngine
using SymEngine: libsymengine, get_symengine_class, is_symbol, Basic, BasicType

adiff(b1::Basic, b2::Basic) = adiff(b1, BasicType(b2))
adiff(b1::Basic, b2::BasicType) =
    throw(ArgumentError("Second argument must be of Symbol type"))

function adiff(b1::Basic, b2::BasicType{Val{:Symbol}})
    a = Basic()
    ret = ccall((:basic_diff, libsymengine), Int, (Ref{Basic}, Ref{Basic}, Ref{Basic}), a, b1, b2)
    return a
end


function bdiff(b1::Basic, b2::Basic)
    is_symbol(b2) || throw(ArgumentError("Second argument must be of Symbol type"))
    a = Basic()
    ret = ccall((:basic_diff, libsymengine), Int, (Ref{Basic}, Ref{Basic}, Ref{Basic}), a, b1, b2)
    return a
end

@vars x
u = sin(x)

julia> @time adiff(u,x)
  0.000023 seconds (3 allocations: 48 bytes)

julia> @time bdiff(u,x)
  0.000017 seconds (2 allocations: 32 bytes)

Compared to the older get_symengine_class we have

julia> @time adiff(u, x)
  0.000024 seconds (4 allocations: 72 bytes)