fix(terminal): send modified keys to the child pty correctly

Keys forwarded into an embedded terminal lost part of their chord, so a
program running there could not tell several distinct keys apart.

`key_to_pty_bytes` tested modifiers one at a time in an if/else-if chain
per key, which collapsed every combination onto its first matching branch:
Ctrl+Shift+Right reached the child as Ctrl+Right (`CSI 1;5C`, not `1;6`).
Home/End/PageUp/PageDown accepted only Ctrl, so Shift+Home arrived bare.
`KeyCode::F(n)` ignored modifiers outright — Shift+F3 arrived as plain F3,
the outgoing-side twin of #699 — and dropped F13 and up. Alt was lost on
every key with no parameterised sequence to carry it (Alt+Backspace, the
readline delete-previous-word, arrived as a bare Backspace), Ctrl+Alt+key
matched neither the Ctrl nor the Alt branch and fell through to the
plain-character arm, and Alt + a non-ASCII character was truncated by an
`as u8` cast.

Encode the modifiers additively in one place instead, as xterm's
`1 + shift|alt<<1|ctrl<<2|meta<<3` parameter, and build each sequence from
shared helpers so a key cannot grow its own divergent chain again. Super
and Hyper have no legacy encoding and are dropped; the kitty protocol is
where they would survive, and the emulator does not speak it to the child.

Established encodings are unchanged — the pre-existing tests pass
untouched, and the new ones all fail against the old encoder.

Author: claude