Different `length` implementations for strings in LLVM/JS

Inspired by [this blog post](https://hsivonen.fi/string-length/) on string lengths, `"🤦🏼‍♂️".length` returns 7 or 17 depending on using LLVM/JS as backend.

We should choose a unified implementation instead - probably UTF-8 everywhere?

Start by locating the string-length implementations in the LLVM and JavaScript backends, then read the linked blog post to understand the competing interpretations. Define which length semantics should be shared, implement the unified behavior in both backends, and verify that the example produces the same result everywhere.

Written by the indexing model from the issue text.