… seq

Three low-severity items.

 - tuish_clear_region re-implements _tuish_str_repeat's O(log n) doubling.
It can't be deduplicated: that helper lives in str.sh, but term.sh depends
only on tui.sh and is sourced without str.sh in minimal setups (README
quick-start, examples/slow_menu.sh). Add a comment documenting the
deliberate duplication so it isn't re-flagged.

 - Conjoining Hangul Jamo (U+1160-U+11FF, medial vowels and final
consonants) were classified width 1; like other wcwidth implementations
they are zero width (they combine onto the leading consonant, which stays
width 2). Add a 0-width branch for that range. New width test: a
decomposed syllable (U+1100 + U+1161 + U+11A8) is 2 columns, not 4.

 - More than cosmetic: tuish_save_cursor used '\x1b7', which ksh93's
printf reads as hex 0x1b7 — so DECSC was emitted as the bytes c6 b7
instead of ESC 7. No single backslash escape works everywhere (the octal
'\0337' becomes byte 337 on mksh). Emit a literal ESC byte (_tuish_chr_27
from the ord table) for both save and restore so the trailing digit can't
be swallowed. Verified ESC 7 / ESC 8 on bash, zsh, ksh93, mksh, busybox
(buffered and unbuffered). New tests/unit/test_term.sh guards it per shell.

All unit suites pass on bash/zsh/ksh93/mksh/busybox; boxes/keyboard/editor
integration pass under ksh93 and mksh.