Discrepancy of OP_RAVI_TOSTRING between VM and JIT

[XmiliaH]

As noted in #223 (comment) there is a discrepancy between the VM and JIT for the OP_RAVI_TOSTRING. The VM allows nil, however, the JIT doesn't.
Following code produces an error after the function is compiled:

function test(s: string)
    return s
end

print(test(nil))

ravi.compile(test)

print(test(nil))

[XmiliaH]

The same behavior can be seen for closure.

[snoopcatt]

Yay, really.

It turns out that we don't have to break backwards compatibility with proposed change.
Because it's already broken (between JIT and VM) 🚬

[dibyendumajumdar]

Sorry my bad. I think my original LLVM backend matched the interpreter semantics.

[dibyendumajumdar]

For now I pushed some changes to improve consistency between JIT & VM.

[dibyendumajumdar]

Hi @XmiliaH if you are able to check that the behavior is consistent now please let me know.