File: src/input.zig:357-362 (advanceOscEsc);
cf. correct pattern at src/input.zig:489-503 (dcs_sixel_esc)
Severity: MEDIUM
Status: ✅ FIXED
advanceOscEsc now aborts the OSC and reprocesses the pending ESC plus the
offending byte through advanceEsc (a bare new ESC keeps the .esc state
pending), so ESC]0;t ESC[31m runs the SGR instead of printing 31m
literally. CAN/SUB-in-OSC fidelity remains tracked under #388.
Test: stray ESC in OSC aborts string and next sequence still runs — bug
#377 (asserts exactly one cell written, with red foreground applied).
For ESC ] 0 ; title ESC [ 3 1 m: after the ESC, [ ≠ \ → toGround() and
the [ is consumed as part of the abort. 31m then renders as literal text.
xterm aborts the string and processes the new escape sequence normally.
Minor fidelity gap also noted by the sweep: CAN/SUB inside OSC strings are
absorbed as content instead of aborting per VT spec.
Mirror dcs_sixel_esc: keep the offending byte and push it back through
feed after leaving OSC; treat CAN (0x18)/SUB (0x1A) as aborts too.
Test: unterminated title followed by SGR — expect colour change, no literal.