Foreign effect ID scanning uses the wrong line-continuation rules #1604

I hit this on main (234cb9b) while returning a constructor from a foreign effect. The source scanner uses C backslash continuation rules for JS comments, but misses C continuations when the file uses CRLF.

For an effect returning Maybe<&1, U32>, this valid JS fails at runtime with ReferenceError: CID is not defined:

function probe() {
  // A path ending in backslash: C:\
  return { $: CID(Some), value: 42 };
}
io_eff(CID(probe), probe);

With CRLF line endings, this valid C comment instead makes compilation fail with CID(NotASymbol) names no constructor or def:

// A quoted macro example: \
CID(NotASymbol)

The C preprocessor keeps the second line in the comment. JS does not: its comment ends at the first newline even when preceded by a backslash. The scanner currently gets both cases backwards. C also allows a CID token itself to be split with backslash-newline; that remains unreplaced and clang rejects it.

I have a small fix that splices C lines before scanning and leaves JS lines alone, plus a regression exercising real constructor returns on both backends. AI-assisted investigation and patch.