fix(lexer): strip backslash-newline line continuations in interpolating strings

[tannevaled]

Bug

Inside a double-quoted / interpolating string literal, a backslash placed immediately before a newline is a line continuation — MRI removes both the backslash and the newline. The escape decoder fell through to its default branch and emitted the raw newline byte, so:

s = "a\
b"
p s      # MRI => "ab"     was => "a\nb"  (WRONG)

Fix

Add the continuation branch to scanStringSegment (the interpolating-string escape decoder that the double-quoted, %Q/%(), and interpolating-heredoc paths all re-lex through). Both \<LF> and \<CR><LF> are stripped entirely; a lone \<CR> stays dropped-backslash + literal CR, matching MRI. The non-interpolating flavours (single-quoted, %q, <<'TERM' heredocs) never reach this decoder and keep the backslash+newline verbatim — also matching MRI.

MRI-match matrix (verified byte-for-byte vs ruby 4.0.5)

flavour	\+newline	changed?
"..." double-quoted	stripped (LF and CRLF)	yes
%Q() / %()	stripped	yes
interpolating heredoc <<~/<<-/<<	stripped	yes
lone \<CR> (no LF)	backslash dropped, CR kept	matches (unchanged)
'...' single-quoted	literal backslash+newline	no
%q()	literal backslash+newline	no
<<'TERM' non-interpolating heredoc	literal backslash+newline	no

Covers continuation-then-escape (\<LF>\t), at end-of-literal, and immediately before #{…}.

Gate

• New tests for every branch; merged coverage 100.0% (CI gate).
• go vet clean; builds on all 6 64-bit arches (amd64/arm64/riscv64/loong64/ppc64le/s390x) + js/wasm + wasip1/wasm.

🤖 Generated with Claude Code