feat(line-editor): UTF-8-aware input, cursor movement, and deletion

What

The line editor was byte-oriented end to end, so multibyte UTF-8 (accented Latin, CJK, emoji) was broken three ways:

• Input: bytes ≥ 0x80 fell through the key-dispatch else arm and were dropped — you couldn't even type café.
• Cursor: Left/Right moved one byte, landing inside a codepoint.
• Deletion: backspace/deleteChar removed one byte → dangling continuation byte (corrupt text); cursor reposition counted bytes not display columns, over-shooting past multibyte chars.

Adds UTF-8 helpers (isUtf8Continuation, utf8SeqLen, prevCodepointLen, curCodepointLen) and:

• insertUtf8(lead) assembles a full codepoint (reading continuation bytes) and inserts it atomically; wired via a 0xC2...0xF4 dispatch arm.
• moveCursorLeft/Right step one whole codepoint.
• backspace/deleteChar remove the whole codepoint at/before the cursor.
• insert/backspace/delete reposition the cursor by display columns (one per codepoint), not bytes.

ASCII is unchanged (every helper returns 1 for ASCII bytes).

Verification

• zig build clean (exit 0).
• PTY: typing echo café prints café; one Backspace → caf; with echo Xé, Left+Left then A → AXé; ASCII regression echo abXcd → mid-line backspace → aXcd still works.

Scope: byte-vs-codepoint correctness. Wide-character (CJK/emoji width-2) column alignment — treating one codepoint as two columns — is a separate follow-up; this treats one codepoint as one column, correct for accented Latin and the common cases.