Summary

The string-literal decoder's catch-all branch appended the byte after \\ verbatim. Three findings flowed from that one gap:

"\q" silently decoded to "q" (A2)
"\0" silently decoded to "0" (A3)
"A" silently decoded to "u0041" (B8)

The char scanner at canonical_text.c:220-222 already whitelists the known escapes and rejects everything else with PAR100. This PR mirrors that policy in the string decoder: explicitly accept n, r, t, and x..; pass through the self-escaped \\, ', "; reject everything else with PAR100. \\x00 continues to be rejected (a NUL byte would truncate a C string), which now matches \\0.

Closes three items from #318.

Test plan

conformance/native/pass/string-escape-canonical.0 covers \\n, \\t, \\", \\\\, \\x41
conformance/native/fail/string-unknown-escape.0 covers "\q"
conformance/native/fail/string-null-escape.0 covers "\0"
pnpm run conformance passes
compiler-metrics.mts has 0 violations (budget bumped by 17)
No C compiler warnings
No existing *.0 source uses "\0" or "\u..." so the rejection is safe