fix(compiler): wrap RHS shifting value by maximum size of type in bits for small integers

[MaxGraey]

Similar to #1494 we should wrap RHS shifting value by maximum size of type in bits.

For booleans we just return leftExpr due to leftExpr <<>> rightExpr & (1 - 1) -> leftExpr <<>> 0 -> leftExpr

• I've read the contributing guidelines

[dcodeIO]

The comment there reads // Cares about garbage bits on the RHS, but only for types smaller than 5 bits (4 bits if i16, 3 bits if i8), making me wonder why we'd want to mask these in addition to relying on that? Might well be that I misunderstood something there.

[MaxGraey]

We need this masks because otherwise it interpret as UB. WebAssembly always use masks for i32/i64 shifts. So we need do the same for emulated types as well. Regarding other langs. C++ don't do this but it's well-known and documented UB. Rust for example always do this (and for small integers as well)

[dcodeIO]

Ah, I see, so this interpolates shift semantics as if there'd be actual i16.shl and i8.shl instructions with 4/3 relevant bits. Should probably document this, but otherwise looks good to do.

[MaxGraey]

You probably better person to properly document this. So you're free to do in this PR or later)