Cache Invocable and prevent concurrent access into orc

[PhilippGrulich]

Whenever invoked, the CallableFunction wrapper locates the executable within the LLVM context.
This makes calling a function not thread-safe as the underlying LLVM implementation is not thread-safe and introduces an unnecessary overhead.

For small functions, this lookup dominates the runtime. Note: This benchmark only measures the invocation. Compilation is done beforehand.

static nautilus::val<int> smallFunction(nautilus::val<int> i) {
	return i * 2;
}

----------------------------------------------------------------------
Benchmark                            Time             CPU   Iterations
----------------------------------------------------------------------
BM_smallFunctionCompiler          1187 ns         1183 ns       565135
BM_smallFunctionInterpreter       62.7 ns         62.5 ns     11918221

This PR caches the Invocable produced by the Compiler during construction of the CallableFunction and thus skips the lookup within the invocation.

----------------------------------------------------------------------
Benchmark                            Time             CPU   Iterations
----------------------------------------------------------------------
BM_smallFunctionCompiler          3.65 ns         3.65 ns    191092600
BM_smallFunctionInterpreter       63.1 ns         63.1 ns     11229368

Note: we can remove the release mode and benchmark commit and only pick the relevant commit